Skip to content

Version History

  • Changes

    Read the outcomes and fixes in this release.

  • Upgrade

    Check migration guidance before changing an existing deployment.

Check release changes before choosing an upgrade.

Find the relevant version below and check breaking changes before scheduling an upgrade.

Next: Upgrade guidance

v3.3.5 (2026-10-26) Latest

Access and log rule conditions now match as written: conditions with dotted values, such as e-mail addresses or IP prefixes, and conditions in outer parentheses joined with and or or could match other requests than written, so deny rules could miss and allow rules could admit identities they were meant to exclude. Two changes may need action, both under Breaking: a parenthesised rule condition that reads both as one comparison and as several now stops startup with an error naming the rule, and REST, GraphQL and MCP search filters such as (cn=x or sn=y) are now read as two comparisons. Bearer sign-in keeps working while the identity provider’s key endpoint is unavailable, forwarded LDAP sign-ins keep their own directory account after a lost directory connection, and wrong-issuer errors stay readable. Set database passwords through database.password or POSTGRES_PASSWORD rather than inside a connection address, where they can reach debug logs; build and documentation tooling is also updated for known advisories.

Breaking

  • Rule conditions in outer parentheses match as written. BREAKING: An access or log rule condition wrapped in outer parentheses and joined with and or or, such as a deny rule (subject.principal.id=alice@example.com or subject.principal.id=bob@example.com), was read as a single comparison with the rest of the condition as its value. With = such a condition never matched, so deny rules did not apply and allow rules admitted no one; with >=, <= or ~= it could match far more than written, so an allow rule such as (subject.principal.id>=alice and subject.principal.id<b) admitted identities like carol.white@example.com that it was meant to exclude. These conditions now work as written. Search filters sent over REST, GraphQL or MCP follow the same rule, so (cn=x or sn=y) now searches for either comparison; to search for a value that contains or followed by =, write the = as \3d. Filters sent by LDAP clients are unchanged. A rule condition that reads both as one comparison and as several comparisons joined by and, or, &&, || or a space, such as (subject.principal.id=alice or subject.principal.id eq bob) or (subject.principal.id>=a&&subject.principal.id<b), now stops startup with an error naming the rule. Migration: remove the outer parentheses, or, for a value that really contains such text, escape the characters that separate it, such as each space as \20. Values that cannot be read as several comparisons, such as (subject.principal.id=uid=alice,ou=Research and Development,dc=example,dc=com), keep matching as written.

Security

  • Rule conditions preserve dotted values. Since 3.0.0, access and log rules with dotted or underscored values, such as e-mail addresses, IP prefixes and operation names, could compare against a rewritten value when using a word operator (eq, startswith, contains, etc.) or spaces around the operator. Such conditions could match too few or too many requests: deny rules did not apply, allow rules did not match (including the shipped example request.client.ip startswith 10.0.), and allow rules with a negated operator such as ne or notin admitted the identities they were meant to exclude. Values are now kept as written. Workaround before upgrading: use a symbol operator without spaces, such as subject.principal.id=alice.smith@example.com or request.client.ip^=10.0.; replace in/oneof lists with one rule per value; for notin, join symbol inequalities with and in the same rule, such as subject.principal.id!=alice.smith@example.com and subject.principal.id!=bob.jones@example.com. Write these FleX conditions without outer parentheses; before this release a parenthesised or can silently never match. For a value that contains spaces, use the LDAP form and write each space as \20, such as where = "(subject.principal.id=alice\\20smith.jones)".
  • Wrong issuer errors stay readable. When a discovery document names a different issuer than configured, Scribe keeps the actual and expected values in the error while escaping control characters and shortening an unusually long actual value.
  • Keep database passwords out of connection addresses. Database passwords given inside database.url, database.host or POSTGRES_HOST can appear in debug logs of the connection pool; set them through database.password or POSTGRES_PASSWORD instead. 3.4 removes them from logs.

Fixed

  • Forwarded LDAP sign-ins keep their own directory account. When a client signed in through a SASL bind, or through any bind with channels.ldap.auth.enabled = false, and Scribe then lost its directory connection for that client, Scribe opened a new connection with its own backend account and ran the client’s later requests there. Later requests that need that directory connection now get UNAVAILABLE until the client binds again, and a write is still never sent twice. Sessions signed in by Scribe itself reconnect as before. See Proxy Authorization.
  • Bearer sign-in survives key-endpoint outages. Bearer tokens signed with the last good keys keep validating when the identity provider’s key endpoint is unavailable or its answer cannot be read, instead of failing after about 20 minutes with the default refresh interval. Entries Scribe cannot read are ignored when the answer also contains a usable signing key. Failed key fetches keep their retry delay. Valid answers retire withdrawn keys within four refresh intervals after the last answer containing them, plus any pending retry delay (up to 60 seconds), including removal of all signing keys and longer cache lifetimes.

v3.3.4 (2026-10-22)

This release fixes one startup failure in 3.3.3 and updates build and documentation tooling for known advisories. If you skipped a startup step to work around it, read the action under Fixed before you start 3.3.4.

Fixed

  • Startup no longer stops while cleaning up entry histories. When startup had to clean up entry histories that earlier resyncs had duplicated, Scribe could stop with Unable to discover event-stream normalization candidates, caused by Cannot change transaction read-only property in the middle of a transaction. We reproduced this failure: a new database connection reached Scribe with a transaction still open. New connections now start without one, which fixes the reproduced failure. Installations whose next start runs this cleanup, mostly the first start after an upgrade, may be affected. Action: if you skipped the cleanup by setting eventStreamNormalizationVersion in the subscriptions table, remove that marker right before the first start of 3.3.4, so the cleanup runs on that start:

    BEGIN;
    UPDATE subscriptions
    SET state = state - 'eventStreamNormalizationVersion'
    WHERE id LIKE 'identity-scribe:state:%';
    SELECT id, state ? 'eventStreamNormalizationVersion' AS marker_present
    FROM subscriptions WHERE id LIKE 'identity-scribe:state:%';
    COMMIT;

    Every row should show marker_present = false. After that start, the log reports the cleanup for each entry type and the marker is set again. Without the marker workaround, no action is needed.

v3.3.3 (2026-10-18)

Bearer tokens and token claims no longer reach logs or traces, access rules match their values literally, and a stopped LDAP listener now restarts on its own; upgrades with renamed entries finish, very large groups load again while oversized entries are refused when written, and two quick changes to an entry keep the newer value. Four changes need action, listed under Breaking: heap dumps need a directory only the service account can read, Observe actions sent from other websites are refused, HTTPS sessions need the protected session cookie, and identity providers must report their configured issuer; bearer sign-in also uses the auth.http trust store now. Logs written by earlier releases may contain tokens and personal data, so delete or restrict them.

Breaking

  • Heap dumps need a private directory. BREAKING: POST /observe/memory/heap-dump now refuses with 503 unless monitoring.memory.heap-dump.directory is owned by the service account and grants no access to anyone else, because a dump holds credentials and personal data; point the setting at such a directory or let Scribe create it. Filesystems without POSIX permissions or ACLs are refused as well.
  • Other websites cannot trigger Observe actions. BREAKING: Inbox actions, the plan-mode kill switch and heap dumps now answer 403 and change nothing when a browser session sends them from another website, including another host of the same domain, or without origin information, so a page elsewhere can no longer act with a signed-in operator’s rights; Inbox actions and the kill switch also answer 415 unless the body is sent with Content-Type: application/json. The Observe pages keep working. All 3.x releases up to and including 3.3.2 are affected; see Sign-in form protection. Migration: switch scripts that call these actions with a session cookie to a bearer token or HTTP Basic credentials, which are not affected, and send JSON bodies with Content-Type: application/json.
  • HTTPS sessions need the protected session cookie. BREAKING: Over HTTPS, Scribe now reads only the session cookie it sets over HTTPS, which no other host and no plain-HTTP page can set; a session cookie left over from a plain-HTTP sign-in, or one hidden inside another cookie’s value, is ignored. Behind a reverse proxy that terminates TLS, a request counts as HTTPS when the proxy is listed in http.requested-uri-discovery and reports https; Scribe then sets the protected cookie with Secure, and access rules on request.secure match. Migration: sign in again after moving a deployment from HTTP to HTTPS; behind a TLS-terminating proxy, enable http.requested-uri-discovery with the proxy as a trusted proxy so sessions get the protected cookie.
  • Identity providers must report their configured issuer — BREAKING: Scribe now rejects a provider whose discovery document names an issuer other than its issuer or public-issuer — bearer tokens, password sign-in and token introspection through that provider fail. Migration: set public-issuer to the issuer your identity provider reports. See Authentication.

Security

  • Access rules match values literally. Underscores, percent signs and backslashes in a starts-with, ends-with, contains or wildcard value acted as wildcards, so a rule could match request paths, hosts or accounts it did not name; they now match only themselves, and only * in a wildcard value matches any text, as documented. All 3.x releases up to and including 3.3.2 are affected; see Access rules.
  • Bearer tokens kept out of logs. A rejected LDAP bearer bind could write the client’s complete bearer token into the operation log, where anyone with log access could reuse it; the log now records only the reason. Delete or restrict logs from earlier releases, and treat tokens that were still valid at the time as exposed.
  • Token claims kept out of logs. A token rejected by a claim check, such as issuer, expiry or audience, wrote its decoded claims, including subject and e-mail address, into the operation log, and introspection failures logged audience values; the log now names the failed checks and claims, the token’s issuer and the expected value, and no claim values. With SCRIBE_LOG_JOSE4J = debug, token headers no longer reach the library log. Delete or restrict logs from earlier releases according to your personal-data retention rules.
  • Rejected opaque tokens kept out of traces. With a UserInfo endpoint configured, an opaque token containing characters not allowed in an HTTP header was written in full into the trace and the operation log; both now record only the reason.
  • Bearer sign-in honours identity provider settings. Bearer tokens are now checked with signing keys and provider details fetched through the auth.http trust store and timeouts, so identity providers with a private certificate authority work without changing the system trust store. While the identity provider is unreachable, Scribe retries it at most every few seconds, backing off to once a minute, instead of on every request, and keeps using the last signing keys it fetched. A provider’s Cache-Control: max-age now counts as at least one minute and at most one day, and provider answers larger than 512 KiB are refused. A failed token request now logs only the provider’s status, the kind of answer (such as JSON or HTML) and its size, no longer the start of its answer or its full content type, which could echo a password or authorization code.

Changed

  • Token rejections report their actual reason. Rejection codes no longer depend on words inside the token: expired tokens report AUTH_TOKEN_EXPIRED, tokens for another audience AUTH_TOKEN_INVALID_AUDIENCE, and a token with an unknown signing algorithm AUTH_TOKEN_INVALID_SIGNATURE with LDAP result 49 and HTTP 401 instead of AUTH_JWKS_ERROR with 52 and 503. All other rejections keep their LDAP result and HTTP status.

Fixed

  • Upgrades with renamed entries finish. An upgrade that rebuilds stored entries failed on every start when an entry had been renamed, including entries deleted later; renamed entries now come back with their new name and last change time.
  • Groups with very many members are stored and read again. An entry with more than 65,536 values in one attribute, such as a large group’s members, was stored but then failed to load, so later changes to it, directory reads of it and restores failed; such entries, including those stored by earlier releases, now load normally. An entry beyond the supported size (more than 4,096 attributes, more than about four million values, an attribute name over 1,024 bytes, a value over 16 MiB or 64 MiB of values in total) is now refused when it is written, with an error naming the exceeded limit, and the previously stored version and its history stay unchanged.
  • /observe/health/ready fails while searches are stalled. Like /readyz, it and the combined /observe/health now return 503 with a query-stall check while every query permit stays in use, unless monitoring.readiness.fail-on-query-stall is false; see Kubernetes probes.
  • Stopped LDAP listeners restart. When an error such as running out of threads stopped an LDAP or IdentityHub listener, it stayed down until Scribe was restarted while the channel reported healthy; it now restarts as set by retry.services, and the channel reports unhealthy until it accepts connections again. On Linux with use-reuse-address = false, the restart retries while an old connection still holds the port and logs ldap_listener_rebind_failed for each attempt.
  • Shutdown no longer waits for stalled clients. Stopping Scribe or a listener no longer waits for a client that has stopped reading its responses.
  • Reconciliation counts only recorded entries. When the database could not record that a page of entries was verified, reconciliation still counted the page and searched the same entries again; it now retries briefly and otherwise stops the run without counting that page.
  • A slow directory no longer ties up database connections. When a change notification arrived without the entry’s object classes, Scribe re-read the entry from the directory while holding a database connection and the entry’s row lock; it now holds them only for the write.
  • Two quick changes keep the newer value. When the directory changed the same entry twice in quick succession, Scribe could store the older values last and add a change back to them to the entry’s history, although the directory still held the newer values; the newer values now stay stored and the history shows no such change. Entries already affected are corrected by their next change or a resync.
  • Later queries keep their settings. A session setting that failed to apply or restore could change plan choice or timeouts for unrelated later queries on the same pooled connection.
  • Transient errors retry under every system locale. Database, LDAP and HTTP operations now also retry temporary I/O failures under Turkish or Azerbaijani system locales.
  • A missing license file is named. When license.file points at a file that does not exist or cannot be read, verification now fails with License file not found or not readable: <path> and where the setting came from.
  • The production template starts. config/production.conf failed startup validation; it now serves the Operator UI and REST admin on the internal socket together with Observe, and all templates note that production needs SCRIBE_AUTH_SESSION_KEYS. See Network isolation.
  • Missing session keys stop startup at once. With auth.enabled = false and browser sessions on, the default when any UI is enabled, a production start without auth.session.cookie-keys now fails validation with AUTH_SESSION_KEYS_REQUIRED instead of retrying until retry.services.max-attempts runs out.
  • The Helm stack serves monitoring on port 9001. The chart exposed port 9001, but Scribe never opened it; the chart now configures health and /observe on 9001, matching the Docker bundle, and its probes check that port.
  • The Helm stack’s Grafana reaches Prometheus and Tempo. The stack now serves Prometheus on 9090 and points the Tempo datasource at its HTTP port, so dashboards and trace lookups work after install.
  • Health endpoints answer HEAD. /livez, /readyz, /startedz and /healthz answered HEAD with 404, so probes such as wget --spider saw a running instance as failed; they now return the same status as GET, without a body.
  • Delegated searches keep the client’s attribute spelling. Since 3.0, searches forwarded to the directory, such as those under ldap.ignore-contexts, sent unmapped attribute names in lowercase, so a client asking for workforceID could receive workforceid and find no values; Scribe now forwards them as the client spells them.
  • Sorted searches by position return each entry once. A sorted search that starts at a numbered position, such as an LDAP virtual list view request by offset, could repeat and skip entries when some had no value for the sort attribute; each position now matches the complete sorted result. All 3.x releases up to and including 3.3.2 are affected.
  • Paged sorted searches return long values once. With the default database.query.sort.long-value-tiebreaker = deterministic, paging through values that share their first 256 characters could repeat some entries and skip others; each page now continues exactly where the previous one ended.
  • Case-only changes replace the old spelling. When the directory changed only the letter case of one value of a multi-valued attribute, Scribe kept both spellings in searchable data and recorded only an addition; it now keeps the new spelling and records the old one as removed. Values recorded before this release are not corrected.
  • Per-transcribe task limit takes effect. The documented max-task-concurrency setting of a transcribe was ignored in favour of the database pool size; it now applies, capped at concurrency. See Transcribes.
  • Browser sign-in works behind a proxied identity provider. When public-issuer differed from issuer, as is usual behind a reverse proxy, every browser sign-in was rejected; it now succeeds, and sign-ins from other identity providers are still refused.
  • Failed MCP tool calls count as failures. A tool call that returned an error, such as an invalid argument, was recorded as ok in the per-tool metrics and the audit log; it is now recorded as unknown, so error rates show.
  • Scheduled maintenance stops at shutdown. Scheduled and interval maintenance runs now stop at their next safe point when Scribe shuts down, as manual runs already did, instead of delaying shutdown until the sweep ends.
  • Entry views show the current state. In the Operator UI, returning to an entry or to a relative time such as “1 hour ago” showed the state from your first visit, and a deleted entry reappeared; the view now checks for the latest state and says when that check failed, with a “Try again” button.
  • Entry history shows the time you chose. In the Operator UI, choosing a time before the entry existed or older than the loaded history jumped to a later change; the view now shows the entry at that time, or says it did not exist then.
  • Entry comparisons show loading and errors. In the Operator UI, comparing two states of an entry showed every value as added or removed while one side was loading or had failed; it now shows that it is loading, or an error with a “Try again” button.
  • The Browse Hub says when its data is missing. When monitoring data failed to load or refresh, Browse Hub cards showed zeros, “None”, “All clear” or old values as if current, also after a reload; they now show “Loading” or “Unavailable”, with a “Try again” button when nothing loaded or a refresh failed.
  • Observe views work with hints off. With monitoring.hints.enabled = false, the Doctor, Services, Build Queue, Ingest, JVM, LDAP Forwarding and Pipeline views showed only an error; they now show their data and mark values that did not refresh as unavailable.
  • The Changes feed shows the right hourly rate. In the Operator UI, the Changes summary rate read 3,600 times too low; 3,600 changes in one hour now show as 3.6k/h.

v3.3.2 (2026-10-13)

Delegated writes recover after directory disconnects, and operation logs now record write durations and outcomes. The bundled compression library also receives a security fix. No configuration or database changes are required.

Security

  • Compression library security update. Updated the compression library bundled with 3.3.1 to address CVE-2026-106451. Existing data remains readable; no configuration changes are required.

Added

  • LDAP writes show their timing. Add, Modify, Delete and rename requests that IdentityScribe forwards to the directory now appear in the operation log as LDAP.Add, LDAP.Modify and LDAP.Delete with their duration and result, and in scribe_channel_request_duration_seconds with op="add", op="modify" or op="delete"; a rename counts as a modify. Each write shows how long the directory took to answer as an LDAP.Delegation event, so a slow write tells whether the time went to IdentityScribe or to the directory, and whether the write first had to connect to the directory. As with searches, slow and failed writes now appear in the operation log by default, except fast ones refused because the entry is missing or the request is invalid; to log every write, see Time LDAP writes. Requests and their results are unchanged.

Fixed

  • Delegated writes recover after disconnects. Creates and modifications in 3.3.0 and 3.3.1 could keep failing with UNAVAILABLE after a directory connection dropped; Scribe now restores the connection for subsequent requests without repeating an uncertain write, and no configuration change is needed.
  • Invalid continuations give clear guidance. Rejected paging or VLV continuations in 3.3.0 and 3.3.1 now say Paging or VLV continuation is not valid on this connection; restart the search. instead of claiming that the authorization identity changed.
  • Malformed paging returns protocol errors. Malformed paged-results request controls in 3.3.0 and 3.3.1 now receive PROTOCOL_ERROR (2) and The paged results request control is malformed. instead of a misleading continuation rejection.

v3.3.1 (2026-10-07)

Changed

  • Documentation explains with figures. Concept, guide, channel, reference and operations pages now show how Scribe works in figures that state their point in the caption, and mark the setting or flag that matters in configuration and command examples. Start with Architecture and Change Detection.

Fixed

  • Upgrade to 3.3 works without superuser — Starting 3.3.0 failed with permission denied to set parameter "identity_scribe.rebuilding_entries" when the database account is not a PostgreSQL superuser, which is the usual case on managed PostgreSQL services. The database upgrade now needs only the rights of the database owner. Installations that already upgraded to 3.3.0 start normally; no action is needed.
  • Query statistics work without superuser — The Observe Queries tab failed to load when the database account is not a PostgreSQL superuser, which is the usual case on managed PostgreSQL services. It now loads and says which step is missing when query statistics are not enabled. On a server shared with other applications it lists the queries of Scribe’s own database account, which others’ queries no longer crowd out. With database.auto-create-query-stats-extension on, Scribe now installs the pg_stat_statements extension when its account is allowed to; on most managed services a database administrator still has to install it.
  • Documentation matches the product. Corrected statements that did not match Scribe’s behaviour, among them: gRPC is listed as planned, not available; the license expiry warnings start 15 days ahead and an expired license stops a running instance at its next check; error responses never carry details; configuration priority, log rules for client errors, task pressure and connection pool sizes are now described as they work.

v3.3.0 (2026-10-03)

Security

  • Oversized HTTP requests are refused early — BREAKING: HTTP requests with a body larger than 1 MiB are now refused with 413 or a closed connection, and sign-in form posts larger than 16 KiB get 413 — clients that send larger GraphQL or MCP requests now fail. Migration: raise http.request-limits.max-payload-size (SCRIBE_HTTP_MAX_PAYLOAD_SIZE) for the socket that needs it, or set it to -1 to remove the limit. Earlier releases accepted HTTP request bodies of any size before sign-in, so a single client could make the server reserve large amounts of memory. See HTTP request limits.
  • Configuration view hides secrets reliably — BREAKING: the configuration view (/observe/config) and hint details requested with redact=false now require Observe write access (by default the admin role), so signed-in accounts without it now get 403; the configuration view and --printconfig now print formatted JSON instead of HOCON. Migration: give operators who need these views the role named in monitoring.observe.write.auth.rules, or adjust those rules; update scripts that parse --printconfig or /observe/config output as HOCON. Previously both showed comments from the configuration files, numeric and true/false values under secret settings, and custom header values such as exporter API keys, so such a secret could be read by any Observe user. They now show no comments and redact every value under a secret setting and every header value, and the view is no longer cached by browsers. The Observe MCP endpoint now follows monitoring.observe.auth.rules; when it runs on its own socket, monitoring.mcp.auth.rules replace them if set. All 3.x releases up to and including 3.2.1 are affected. See Endpoints.
  • LDAP token errors protect private details. Bearer sign-in and proxy authorization now keep token, account lookup, and unexpected validation details out of LDAP responses. If the identity provider cannot supply signing keys, requests receive a temporary unavailable response. Operators can find the cause in service logs.
  • Local account passwords stay local. Alternate directory name spellings now resolve to the configured local account, so an incorrect password cannot be sent to the directory.
  • Sign-in redirects stay within allowed pages. Encoded parent-directory segments can no longer move a sign-in redirect outside its allowed page area.
  • Signing out over HTTPS ends the session. Signing out over HTTPS left the session cookie in the browser, so the session stayed valid there until it expired. Sign-out now removes it. All 3.x releases up to and including 3.2.1 are affected. No configuration change is needed.
  • Other websites cannot change your session — BREAKING: sign-in and sign-out POST requests without origin information are rejected, and plain-HTTP deployments and older browsers such as Safari before 16.4 need the reverse proxy to report the public address — scripts that post the sign-in form without an Origin header, and those deployments and browsers behind a reverse proxy that does not, now get “Invalid request” instead of a session. Migration: switch scripts that post the sign-in form to HTTP Basic; for plain HTTP or older browsers behind a reverse proxy, enable http.requested-uri-discovery, list the connection peer as trusted, and have it send Forwarded with proto and host, or X-Forwarded-For, X-Forwarded-Proto, and X-Forwarded-Host. Another website could submit the sign-in form with an account of its choosing and sign a visitor’s browser into that account without the visitor noticing. The sign-in page now accepts a form only from its own pages; a form posted from elsewhere shows “Invalid request” and creates no session. Over HTTPS, sign-in behind a reverse proxy keeps working in current browsers without configuration changes. Sign-out now rejects cross-site requests and requests without origin information, leaving the current session in place; the in-app sign-out still works, and says so when a sign-out is rejected instead of returning to the sign-in page. Scripts that post the sign-out form without an Origin header now get “Invalid request”. HTTP Basic, OIDC sign-in, and LDAP bind are not affected. All 3.x releases up to and including 3.2.1 are affected. See Sign-in form protection.
  • Proxy authorization now takes effect — BREAKING: an LDAP request with a Proxy Authorization control is now checked, and when IdentityScribe forwards it to the directory it runs as the requested identity instead of the signed-in account — a client that names a DN or username without permission, or presents a bearer token that maps to no directory account, now gets AUTHORIZATION_DENIED (123); a non-critical, repeated, or malformed control gets PROTOCOL_ERROR (2), and a bind with an otherwise valid control gets UNAVAILABLE_CRITICAL_EXTENSION (12); a local account without the permission gets 123 for the control on every operation except a bind, which gets 2 or 12 like any bind; and an abandon request carrying the control is ignored. Migration: list each account that may act for other users in channels.ldap.proxy-authorization.allowed (for the Identity Hub listener, channels.identity-hub.proxy-authorization.allowed) and make sure IdentityScribe’s directory account may act for those users in the directory; set proxy-authorization.enabled = true on a local account that should do the same; remove the control from clients that do not need it and from bind requests. Earlier releases ignored the control, so such requests ran with the signed-in account’s directory rights rather than the requested user’s. IdentityScribe now decides before it looks up the requested identity or contacts the directory: a bearer: token proves its subject and needs no permission where the listener accepts bearer sign-in, while a DN or username needs the new permission, which no account has by default and which a SASL sign-in gets only when the directory reports the signed-in account. Searches, writes, and other operations forwarded to the directory then run as the requested identity, and Who-Am-I reports it; searches answered from IdentityScribe’s own data and channel access rules still use the signed-in account. A paged or virtual list view search continues only as the identity that started it. Local accounts keep rejecting the control unless they opt in. All 3.x releases up to and including 3.2.1 are affected. See Proxy Authorization.
  • SASL binds are off by default — BREAKING: the LDAP listeners now refuse every SASL bind with AUTH_METHOD_NOT_SUPPORTED (7) unless its mechanism is listed, and with channel authentication on a listed mechanism counts as a directory sign-in: it needs ldap in the listener’s auth.methods, the account the directory reports through Who-Am-I (RFC 4532) must lie under auth.ldap.base and match auth.ldap.filter and must not be a local account’s DN, its roles come from auth.ldap.roles, SASL binds that cannot reach the directory count toward detecting an outage, and while directory sign-ins fail fast during an outage a new SASL bind fails fast too, unless it is the sign-in let through to check the directory, which IdentityScribe checks before forwarding the bind — a SASL bind whose account the directory does not report, or reports outside that base or filter or as a local account’s DN, fails with INVALID_CREDENTIALS (49). Migration: SASL binds through IdentityScribe are off by default; list the mechanisms you use in channels.ldap.sasl.mechanisms (for the Identity Hub listener, channels.identity-hub.sasl.mechanisms), make sure the directory answers Who-Am-I for them, and move clients that use EXTERNAL, which can no longer be enabled, to another mechanism or a simple bind. Earlier releases forwarded every SASL bind and took the account from the name in the bind request, which SASL does not check, so a client could sign in with its own credentials and be treated as another account; an EXTERNAL bind authenticated IdentityScribe’s own directory certificate. A paged or virtual list view search started while a multi-step SASL bind was in progress no longer continues after the bind completes. All 3.x releases up to and including 3.2.1 are affected. See SASL binds.
  • LDAP sign-ins need a directory account — BREAKING: an LDAP bind that names no directory account now fails with INVALID_CREDENTIALS (49): a dn: bind whose value is not a DN, and a bearer or u: bind whose token or username maps to no DN — such clients can no longer sign in. Migration: bind with the account’s full DN (dn:uid=alice,ou=people,dc=example,dc=com) or its username (u:alice or the plain name), and configure auth.ldap so bearer and username binds find the account’s DN. Earlier releases signed such a client in without an account, and requests IdentityScribe forwarded to the directory for it ran with IdentityScribe’s own directory rights. All 3.x releases up to and including 3.2.1 are affected. See Authentication.
  • Refused LDAP binds leave nothing signed in — With channels.ldap.auth.enabled = false, a bind the directory accepted but whose account IdentityScribe could not determine, such as an Active Directory user principal name, was refused, yet the client’s later requests on that connection still ran with that account’s directory rights. They now run as an anonymous client, as after any refused bind. All 3.x releases up to and including 3.2.1 are affected.
  • Oversized LDAP requests are refused — BREAKING: the LDAP listeners now disconnect a client that sends a request larger than 10 MiB — a client that writes larger entries or modifications through IdentityScribe now loses its connection. Migration: if your clients send larger requests and the directory accepts them, raise channels.ldap.max-message-size-bytes (for the Identity Hub listener, channels.identity-hub.max-message-size-bytes), or set it to 0 to remove the limit. Earlier releases accepted requests of any size, from clients that had not signed in too. Passwords and SASL credentials larger than 64 KiB, bind DNs and usernames larger than 8 KiB, and bind request controls larger than 8 KiB in total now get INVALID_CREDENTIALS (49) before IdentityScribe contacts the directory, which would drop such a request; Kerberos tokens and bearer: tokens up to 64 KiB still fit. All 3.x releases up to and including 3.2.1 are affected. See LDAP Channel.
  • JSON parsing receives security fixes. Updated the bundled JSON library with fixes for denial-of-service and deserialization issues. No configuration change is needed.
  • Web server receives security hardening. Updated the web server with security fixes for request handling. No configuration change is needed.

Added

  • Sign-ins fail fast during directory outages — When the directory stops answering or repeatedly returns busy or unavailable, LDAP sign-ins that need it no longer wait for the ten-second timeout. After five failed sign-in checks in a row, IdentityScribe checks the directory itself; when that check fails too, sign-ins fail immediately with LDAP result 52, HTTP 503, or “Sign-in temporarily unavailable” on the login page, and every ten seconds the next sign-in first checks whether the directory answers again. Sign-ins that fail because of what a client sent, such as wrong passwords or credentials larger than the directory accepts, never make other sign-ins fail fast. Normal sign-ins resume without a restart. Sign-ins rejected this way are counted instead of logged one by one, so clients that retry during an outage do not flood the log. Local accounts, browser sessions, cached sign-ins, and searches are not affected. Adjust or turn this off with auth.ldap.breaker.threshold, auth.ldap.breaker.open-duration, and auth.ldap.breaker.enabled; see Authentication.
  • Doctor reports directory sign-in outages — The new auth.ldap.upstream doctor check and the scribe_ldap_upstream_state gauge show whether the directory answers sign-ins: degraded when sign-in checks fail more often than usual, critical while sign-ins fail fast. Readiness is intentionally unchanged, because the directory serves every instance and taking them all out of rotation would not help. See Monitoring.
  • Release archives include third-party attribution. The release archive now contains THIRD_PARTY_LICENSES.md with the notices and license texts for its open-source components.
  • Readiness reports stalled searches — /readyz and /healthz now return 503 while searches are stalled: every query permit is in use, a search has waited for a permit or was turned away since the last release, and no permit has been released for 60 seconds. A turned-away search still counts after it gave up. A stalled instance usually needs a restart, so an orchestrator can now take it out of rotation; if a search finishes after all, the stall ends and the probes return 200 without one. A busy instance whose searches keep finishing stays ready, and /livez is unchanged. The new query.permit_stall doctor check and scribe_query_permit_stall_seconds gauge report the stall, and the log records a query_permit_stall warning when it starts and query_permit_stall_cleared when it ends. If you restart, collect a thread dump with kill -3 first. Tune the wait with monitoring.readiness.query-stall-after (SCRIBE_MONITORING_READINESS_QUERY_STALL_AFTER), or keep readiness independent of searches with monitoring.readiness.fail-on-query-stall = false (SCRIBE_MONITORING_READINESS_FAIL_ON_QUERY_STALL=false).

Changed

  • LDAP StartTLS setting now rejected — BREAKING: IdentityScribe refuses to start when start-tls or ssl.insecure is set to true for an LDAP connection — configurations with either setting in any LDAP block now fail at startup. Migration: switch to ldaps:// (usually port 636) and set ldap.ssl.ca if your directory uses a private CA. The production guide previously suggested ldap.start-tls = true, but this setting was never applied, so the directory connection stayed unencrypted. See Production guide.
  • Faster planning choices adopted more often. Repeated searches switch to a consistently faster planning choice more often when response times vary. A choice that is slower for typical searches is no longer adopted only because its slowest searches finish sooner.
  • Environment variables override the configuration file — BREAKING: environment variables listed in the configuration reference now take precedence over values in the configuration file — a stale variable left in a service unit or container environment now wins over the file, for example pointing the service at an old database or directory, and a variable the service previously ignored or read leniently now stops startup when invalid, such as an OTLP exporter timeout, an on/off value like 1 or TRUE, or an invalid duration or number; an invalid OTLP protocol stops startup when that exporter is enabled. Migration: before upgrading, list the environment variables the service receives from service units, container definitions, and shell profiles; remove any that should not apply; use true, false, yes, no, on, or off for on/off settings and valid durations, numbers, and OTLP settings; after upgrading, check the config_env_override startup warning, which names the settings an environment variable changed without showing values. Empty variables count as unset, and list settings keep their comma- or space-separated form, such as SCRIBE_AUTH_METHODS=bearer,ldap. Object settings such as SCRIBE_LDAP_VIRTUAL_ATTRIBUTES now accept a value like {} even without a file value. SCRIBE_LICENSE_FILE and the POSTGRES_HOST, POSTGRES_PORT, POSTGRES_USER, and POSTGRES_PASSWORD fallbacks are unchanged: a value in the configuration file still wins. The --license command-line option and a license.file Java system property now take precedence over license.file in the configuration file; previously the file value won, so remove a stale --license from service units before upgrading. See configuration precedence and upgrading to v3.3.0.

Fixed

  • Startup failures consistently signal failure — IdentityScribe now always ends with exit code 1 when it gives up starting, including when a required setting such as auth.session.cookie-keys is missing and every retry fails; this case used to end with 0. Docker restart policies, systemd and Kubernetes now see such a start as failed and can restart or alert. An HTTP port that is already in use now also fails startup: IdentityScribe retries according to retry.services and ends with exit code 1 if those attempts are exhausted, instead of running without health checks, metrics, APIs or the admin UI. A requested stop is not affected.
  • Slower planning choices reverted sooner. Recent measurements help repeated searches return sooner to the standard planning choice when an adopted choice becomes slower.
  • Rare search and shutdown hangs fixed — IdentityScribe is now built on Java 25.0.4 (GraalVM Community Edition 25i2). Earlier releases were built on Java 25.0.2, where a task waiting with a timeout could, rarely, miss the signal to continue and wait forever, so a search, sign-in, or shutdown could hang until restart. The release binary contains its Java runtime, so installing this release is all you need to do. If you build IdentityScribe from source, use GraalVM Community Edition 25i2 (Java 25.0.4) or later.
  • Faster totals for broad either-or searches — The exact total of a search for one of several values or prefixes on different attributes, such as (|(givenName=Mara)(cn=Mara)), could take about twice as long as needed when it matched many entries and IdentityScribe had not yet learned how many entries each part matches, and the estimated total stayed at 200 whatever the real count. These totals are now computed the standard way in that case. All 3.x releases up to and including 3.2.1 are affected. No configuration change is needed.
  • Sorted contains searches tune themselves. Contains searches sorted by the searched attribute now record their timings like other searches, so adaptive planning can pick a faster choice for them and Query Diagnostic Reports show them.
  • Backward paging returns the previous page. Fixed a case where paging backward with a before cursor through a search sorted by a single-valued attribute returned entries far from the cursor instead of the previous page, and placed entries without the sort attribute on the wrong side of it. No configuration change is needed.
  • Sorted paging retains missing values. Fixed a case where cursor paging omitted entries without the sort attribute after a page ended on an entry with a value, when moving forward in ascending order or backward in descending order. No configuration change is needed.
  • Grafana monitoring bundle starts reliably — The Docker full stack in the monitoring bundle now starts its PostgreSQL 18 database and serves the monitoring endpoints on port 9001 that the health check and Grafana dashboards use. The bundle reads the LDAP source, the entries to sync, your license file and a session key for the Operator UI from a .env file, and can publish its database on another port when one on your host already uses 5432; see Grafana monitoring bundle for the prerequisites. Building the full-stack image yourself currently needs an x86_64 host. A database volume that holds data from another PostgreSQL major version needs pg_upgrade or a dump and restore.
  • Late deletes preserve replacement entries. A delayed deletion for a removed entry no longer removes a new entry created under the same name.
  • Forwarded requests explain directory outages. If a forwarded LDAP request cannot open a directory connection, it now returns UNAVAILABLE (52) with a plain message. New forwarded binds and writes previously returned 80; a first forwarded search could return a different connection code. Unexpected search failures also keep private details out of the response.
  • Dropped directory connections answer consistently. If the directory connection behind a forwarded LDAP request drops during a session and cannot be reopened, searches, binds, writes and other forwarded operations now return UNAVAILABLE (52) with “The directory is temporarily unavailable.”, the same answer as when no connection can be opened. They previously returned SERVER_DOWN (81) or another client-side code such as 85 or 91, with connection error text that could name the directory host.
  • Error reference shows actual messages. The error reference now lists the messages clients can receive for token, export, lookup, and GraphQL failures, plus a search that fails a consistency check, in place of generic text.
  • Large legacy upgrades finish rebuilding — A startup rebuild when upgrading from 1.x could fail repeatedly on a large directory with the default database settings. These rebuilds now complete without increasing the database lock-memory allowance. Upgrades from 3.2 do not run this rebuild.
  • Shared instances recheck sorted searches — Instances sharing a database now stop using affected sort indexes during startup rebuilding and verify them again afterwards. New installations and upgraded instances resume normal sorted-search performance after verification without false backfill alerts. The convergence view shows when the readiness check is paused instead of reporting progress that it cannot verify. Upgrades from 1.x still require a maintenance window. See Upgrading.
  • Sorted searches with totals succeed and run faster — A sorted search that asked for the total number of matches could fail instead of returning results: LDAP paged searches returned result 19 (constraint violation) with “Can’t do that right now”, REST returned HTTP 409, and GraphQL and MCP returned an error. This affected first pages that request only the total or name a virtual attribute among the requested attributes, for two kinds of filter on the sort attribute: a lower bound such as (sn>=M) sorted by sn, and a prefix such as (cn=Mei*) sorted by cn that matches more than 10,000 entries of one type, from the second such search on. These searches now return their entries with the exact total, and their exact and estimated totals are computed the same way as for the search without sorting, so they take about as long and estimates are as close. All 3.x releases up to and including 3.2.1 are affected. Totals for other sorted searches are faster as well: in our benchmarks on a directory with 250,000 entries, they were typically about three times faster than in 3.2.1, and up to about 20 times faster for broad filters on an attribute other than the sort attribute, such as a “starts with” search on a common first-name prefix sorted by surname. Totals for sorted searches that match only a few entries were already fast and stay that way. No configuration change is needed.
  • Debug logging reaches every log category — --debug, --trace, and a debug or trace log.level or SCRIBE_LOG_LEVEL now also apply to the SuperVisor, Ingest, Monitoring, License, and Config log categories, which stayed at info before. A quieter global level, such as --quiet or log.level = warn, still keeps their operation logs at info. A level set for one category, such as SCRIBE_LOG_MONITORING=warn, still wins in both directions. See Logging.
  • Library log levels can be changed — SCRIBE_LOG_KENOXA, SCRIBE_LOG_HIKARI, SCRIBE_LOG_FLYWAY, SCRIBE_LOG_JOSE4J, and SCRIBE_LOG_HELIDON had no effect, and neither did a quoted key such as log."com.zaxxer.hikari" in the configuration file, so these libraries always logged at warn. The variables now apply. In the configuration file, write the setting without quotes, as in log.com.zaxxer.hikari = debug; settings already written that way keep working. If one of these variables is set in your environment, the library now logs at that level. See Logging.
  • LDAP synchronization settings take effect — ldap.checkpoint-offset, ldap.idle-period, ldap.ignore-checkpoint-timestamp, and ldap.single-value-attributes, set in the configuration file or through SCRIBE_LDAP_CHECKPOINT_OFFSET, SCRIBE_LDAP_IDLE_PERIOD, SCRIBE_LDAP_IGNORE_CHECKPOINT_TIMESTAMP, and SCRIBE_LDAP_SINGLE_VALUE_ATTRIBUTES, were ignored, so synchronization always used the built-in values and the directory schema alone. They now apply. If you set ldap.ignore-checkpoint-timestamp = true, every start now runs a full synchronization; remove it unless that is intended. If you set ldap.single-value-attributes, check that the listed attributes hold one value in the directory. See LDAP.
  • Maintenance and reconciliation thresholds take effect — database.maintenance.soft-thresholds, database.maintenance.hard-thresholds, ldap.reconciliation.soft-thresholds, and ldap.reconciliation.hard-thresholds, set in the configuration file or through their environment variables, were ignored, so maintenance and reconciliation always used the built-in thresholds. They now apply, and a configured set of thresholds replaces the built-in one. If you set them, check that the values still describe how idle the system must be before maintenance or reconciliation runs. See Database and LDAP.
  • Duplicate local account aliases rejected — Two local accounts whose username aliases differed only by surrounding spaces or invisible characters, or an alias that was another account’s DN with such padding, passed configuration validation, and the alias then signed in as nobody. Such a configuration is now refused at startup with a clear message, as is an alias that consists only of padding. Configurations with distinct aliases are not affected.
  • Trusted proxy address ranges now work — BREAKING: http.requested-uri-discovery.trusted-proxies.exact now matches CIDR ranges and IPv6 address spellings by address, and rejects non-IP entries at startup — a host name that was previously ignored can now prevent startup. Migration: replace non-IP entries with IP addresses or CIDR ranges; move any intended pattern over the peer IP address to http.requested-uri-discovery.trusted-proxies.pattern.
  • Reverse proxies report the public address — BREAKING: with trusted X-Forwarded-* headers, IdentityScribe now uses the public address for sign-in origin checks, CORS, OIDC redirect URIs, and OAuth metadata — OIDC sign-in fails at the identity provider when the redirect URI registered there uses the internal or http:// address. Migration: register the public HTTPS redirect URI at the identity provider. See Sign-in form protection.
  • CORS accepts standard web origins. With http.cors.allow-origins = [same-origin], requests to HTTPS port 443 or HTTP port 80 now match the browser’s origin. The host value of http.cors.allow-origins returns the same address without an unnecessary default port.
  • OIDC sign-in works over HTTPS — OIDC sign-in failed with “CSRF validation failed” whenever IdentityScribe received the sign-in over HTTPS, because browsers discarded its temporary verification cookie. The cookie is now accepted and removed after sign-in, and the new session cookie is kept. All 3.x releases up to and including 3.2.1 are affected. No configuration change is needed.
  • MCP lookups reject bad input clearly. Reference searches now ignore surrounding spaces and keep long misses concise. Unknown help topics return a validation error that lists the accepted topics. The hint shown with invalid-argument errors now points to a help topic and a reference search that work.
  • Sign-in and sign-out reach any page. Redirects after sign-in and sign-out now work when the destination page or its query contains non-ASCII characters, such as umlauts.
  • Directory outages show temporary sign-in failure — When the directory cannot be reached or returns busy or unavailable, the sign-in page and every authentication channel now report a temporary failure instead of wrong credentials. LDAP bind returns 52 and HTTP requests return 503. Local account sign-ins remain available during a directory outage.
  • Early stream closes respect concurrency limits — When a client closed a search result stream before reading it to the end, the service could let more searches reach the database at once than the configured maximum, adding load to an already busy instance. Early stream closes now stay within the limit on every channel. All 3.x releases up to and including 3.2.1 are affected, and no configuration change is needed.
  • Numeric local account DNs recognised — A configured local account is now recognised when its login DN is presented with numeric attribute types, such as 2.5.4.3=proxy,o=global for cn=proxy,o=global, on all channels (browser login, HTTP Basic, and LDAP bind). Earlier releases treated that spelling as an unknown name and forwarded it with the submitted password to the directory, so a directory entry at the same DN could sign in even though the DN belongs to a local account. Local accounts were introduced in 3.2.0, so 3.2.0 and 3.2.1 are affected. As with the named form, a local account signs in only with its configured password, and a wrong password is never tried against LDAP. No configuration change is needed.
  • Database environment variables no longer block startup — Setting nested and some other database.* options by environment variable, such as SCRIBE_DATABASE_FILTER_SELECTIVITY_PLAN_MODE_ENABLED=false, SCRIBE_DATABASE_ORPHANED_TRANSCRIBES_BEHAVIOR=ignore, or SCRIBE_DATABASE_QUERY_HTTP_ACQUISITION_TIMEOUT=10s, now takes effect. Previously the service refused to start with an Unsupported property name error. No configuration change is needed.
  • Busy searches no longer disrupt synchronization — Under heavy search load while the database was slow to write, saving query diagnostic report details could occupy every connection that synchronization needs. Searches then waited up to 30 seconds, synchronization logged “Failed to persist entry/event” and re-read entries from its last checkpoint, and the log filled with “Failed to persist query diagnostic snapshot” warnings. Report details are now saved in the background, one query at a time, and only while a connection beyond those synchronization needs is free; otherwise they are skipped instead of waiting, and a save that takes longer than two seconds is abandoned. Synchronization needs one connection per transcribe, at least two. If database.system-pool-size is set to exactly that, report details are not saved and a warning at startup says so; raise it by one to save them. While the database is busy, only the latest details of each query are kept, and if more queries wait to be saved than monitoring.hints.cache.size allows, the others are skipped. Skipped details are not retried automatically; they may still be saved when the current query completes or when the query runs again, if there is room. The log records one warning when skipping starts, and one with the skipped count and the time of the last skip on the first save at least a minute after that skip, or at shutdown if skipping had not ended by then. A database that stays near its limit logs at most one such pair per minute. All 3.x releases up to and including 3.2.1 are affected. No other configuration change is needed.
  • Sorted searches recover after first start — After the first start on a new database, sorted searches could keep using a slower plan until the next restart, and /observe/convergence reported sort indexes as ready while none were being tracked. Sort indexes are now rebuilt during that start and become ready on their own, and the convergence report counts them as backfilling until they are. All 3.x releases up to and including 3.2.1 are affected. No configuration change is needed. See Deployment verification.
  • Adaptive planning stays active continuously. Searches no longer fall back to the standard planning choice at regular intervals while the database responds normally, including during quiet periods. Turning adaptive planning off takes effect promptly even after a clock correction, and still lets searches already running finish.
  • License checks ignore system language — BREAKING: the directory product is now identified the same way under every system language — an installation that runs under a Turkish or Azerbaijani system language with an UnboundID or Ping Identity directory fails to start with Invalid scopes until its license is reissued. Migration: on such an installation, send the license request the upgraded service shows, as described in Renewing a License. Under those system languages, such a directory was identified under a different product name, so a license requested under one system language failed with Invalid scopes when the service started under another. On such an installation, the first start after the upgrade also runs one full synchronization of every entry type. Other installations keep their license and synchronization state.
  • A stop during startup ends cleanly — A stop that arrived just as the service, the synchronization of a transcribe, or an LDAP listener was starting, or restarting after an error, could leave it marked as failed, and the shutdown then logged an error although nothing had started. Such a stop now ends cleanly. No configuration change is needed.
  • Forwarded virtual list view searches work — Every virtual list view (VLV) search that IdentityScribe forwards to the directory, such as a search below ldap.ignore-contexts, failed with “Paging or VLV continuation is invalid after authentication changed; restart the search.” and was logged as an unimplemented search, even when the client had not signed in again. Identity Hub with VLV paging hit this on every forwarded search. These searches are forwarded again, and like other first-page searches they now reconnect once when their directory connection has gone stale. Continuing a paged or VLV result after signing in as another account is still rejected. A VLV request with a malformed control, and any request with a control whose type is not a plain numeric object identifier such as 2.16.840.1.113730.3.4.9, is now answered with a protocol error, and a search rejected this way tells the client why. 3.2.0 and 3.2.1 are affected. No configuration change is needed.
  • Multi-value lookups report correct totals — A search for any of several values of an attribute listed under value-match, such as (|(workforceID=A)(workforceID=B)), returned the matching entries but reported a total of 0 in LDAP paged results and virtual list view responses. Exact and estimated totals for such searches on the REST, GraphQL, and MCP channels were wrong as well. Identity Hub 9.13 and earlier rely on that total in their default paged search and therefore returned no results; later Identity Hub releases returned the entries but could report a total of 0. Such searches now report the number of matching entries. All 3.x releases up to and including 3.2.1 are affected. No configuration change is needed.

v3.2.1 (2026-09-21)

Security

  • Compression libraries receive security updates. Updated both compression libraries to address known vulnerabilities. No configuration changes are required.

Fixed

  • Searches no longer stall until restart — On a busy service, directory searches on every channel could stop returning results until the service was restarted, while synchronization and health checks still reported a healthy instance. This permanent stall can no longer occur. All 3.x releases up to and including 3.2.0 are affected, so rolling back to an earlier 3.x release does not avoid the problem.
  • Searches recover after clients give up or time out — Searches on every channel could stop after clients repeatedly gave up on slow requests or searches ran into their time limit while waiting, leaving a restart as the only recovery. Abandoned requests no longer cause this permanent stall. This affects releases from 3.0.0-rc.3 through 3.2.0, including stable releases 3.0.0, 3.1.0, and 3.2.0.
  • Browser login works for local and LDAP accounts together — With auth.methods = [local, ldap], the login form now accepts both configured local accounts and LDAP users. In 3.2.0, adding a local technical account made browser login local-only and LDAP users were rejected with “Invalid credentials”. No configuration change is needed; if you set auth.session.method = ldap as a workaround, you can keep it to keep technical accounts out of the browser, or remove it to allow both. Wherever local is accepted, a local account signs in only with its configured password, and a wrong password for a local DN or alias is never tried against LDAP. Surrounding whitespace in a login name, including non-breaking and invisible spaces from copy and paste, is now ignored for local accounts on all channels (browser login, HTTP Basic, and LDAP bind), and a local account alias configured with surrounding whitespace matches its trimmed form. [local] and [ldap] on their own, OIDC, ROPC, HTTP Basic, and LDAP bind behave as before.
  • MCP tools name wrongly typed arguments — A Data MCP tool call whose arguments use the wrong JSON types, such as a string limit or an array entryType, now returns an invalid-argument error that names the argument. Earlier releases reported this as a temporary failure.
  • Observe MCP rejects unknown scopes — An observe call whose scope is not one of the documented values now returns an invalid-argument error. The submitted value is truncated and is not copied into the public error text, so a very long scope cannot inflate the response. Earlier releases silently returned the summary.
  • MCP ref lookups now resolve — Asking ref for an error code, metric, configuration key, or attribute now returns the matching documentation URL. Earlier releases returned no match. When nothing matches, the suggested queries are names that do resolve.

v3.2.0 (2026-09-18)

Added

  • Technical accounts can authenticate locally — Configure one or more local accounts with a DN, password, and optional short-name alias for LDAP, HTTP Basic, or browser login without a backend authentication round trip. Forwarded LDAP operations use the account’s explicit backend credentials or the root LDAP credentials as a complete pair, and disabled delegation fails before a backend connection is opened.
  • Thread dumps from a running service — Sending the quit signal (kill -3) now writes every thread’s state and code location to the service output and lets the process carry on serving. Operators can see what a stalled instance is waiting on without restarting it, and the dump holds no directory entries or credentials. Earlier versions shut down on that signal.

Changed

  • Authentication settings keep their method scope — LDAP and ROPC cache lifetimes, disabled caching, and token-expiry revalidation now apply to the selected authentication method. Concurrent checks for the same backend credentials share bounded work; local passwords are never cached, and bearer validation remains unchanged.

Fixed

  • LDAP sign-ins stop waiting indefinitely — Authentication and identity lookups now have a ten-second default timeout, with limited simultaneous upstream work. Configure auth.ldap.request-timeout from 1 millisecond to 1 minute. Brief bursts can wait up to two seconds at each capacity limit within the authentication deadline. The default pending limit is 256 callers, including those joining a check for the same credential. Size auth.ldap.max-pending for the client connections that may bind within one upstream authentication round-trip, such as a client pool reconnecting after a restart; adjust waiting with auth.ldap.capacity-wait. Capacity rejections return LDAP busy (51), and authentication timeouts return unavailable (52); HTTP Basic sign-ins return 503 for either. These temporary failures skip the credential-failure delay. Invalid credentials retain LDAP 49, HTTP 401, and their configured minimum response duration.

v3.1.0 (2026-08-31)

Added

  • Query diagnostics expose each stage — The Query Diagnostic Report now breaks a captured operation’s time into Normalize, Plan, Prepare, and Execute stages and shows how much time preparation used. The share covers one executed search or count operation; Execute includes admission wait, database work, and result delivery.

Breaking

  • Memory pressure metric matches documentation — BREAKING: Prometheus now exports JVM memory pressure as jvm_memory_pressure instead of jvm_memory_pressure_ratio — custom dashboards and alerts that query jvm_memory_pressure_ratio stop receiving samples. Migration: replace jvm_memory_pressure_ratio with jvm_memory_pressure; the bundled dashboards are already updated.

  • Workload evidence separates retained examples — BREAKING: Observe workload responses and support exports now separate aggregate observations from bounded retained examples — clients that read seenCount, card recurrence fields, or samples can no longer parse those removed fields, and workload support exports now use schema 2. Migration: read observedCount, evidenceCoverage, and retainedExamples; use retainedBecause to explain each example; update support-export consumers to schema 2.

  • Affected entry histories repair automatically — BREAKING: On upgrade, histories affected by the former identity reset are repaired automatically when the stored evidence is unambiguous, restoring the original entry ID as the single identity, removing the false Created event, and reducing later changes to the attributes that actually changed — temporary entry IDs created during affected refreshes no longer resolve. Migration: re-resolve affected entries by UUID or distinguished name and refresh any cached entry IDs; ambiguous histories remain unchanged and available for support diagnostics.

Fixed

  • Entry history survives configuration changes — Existing entries retain their entry ID and complete history when synchronized attributes or other source settings are added, changed, or removed and a full refresh follows. Only real object changes create history events; newly configured attributes that are absent on an object and attributes removed from synchronization update the current view without inventing changes. An interrupted or incomplete refresh remains pending instead of deleting entries that were not safely checked.

  • Attribute Coverage lists every attribute — The Operator UI now lists every stored attribute, including attributes without index or sort coverage. Those rows use a neutral “Not applicable” state and do not report healthy coverage signals.

  • Slow shares use every operation — Observe calculates each slow share from the complete workload seen by the current process. Drill-downs keep a bounded typical comparison beside noteworthy examples, while the overview stays focused on actionable patterns. Slow or failed examples no longer act as the denominator for the aggregate.

v3.0.0 (2026-07-31)

Breaking

  • BREAKING: Negative pagination limits now fail consistently across REST, GraphQL, and MCP — GraphQL directory searches that previously used the configured default now fail, GraphQL change-history errors now use ARGUMENT_INVALID_VALUE instead of INVALID_ARGUMENT, and MCP search and change requests now fail instead of requesting one result. Migration: send a positive limit, or omit it to use the configured default.

  • Monitoring log-rule operation shortcuts are no longer accepted through monitoring.log.rules[].name — startup fails if the shortcut remains in a rule. Migration: move exact operation filters to where = "scribe.operation=Exact.Operation" and pattern filters to where = "scribe.operation~=Prefix.*"; keep any duration or result predicates in the same where expression.

    monitoring.log.rules = [
    { action = exclude, name = "Transcription.WorkItem", where = "duration.seconds<=500ms" }
    { action = exclude, where = "scribe.operation=Transcription.WorkItem duration.seconds<=500ms" }
    ]
  • Hint persistence operation shortcuts are no longer accepted through monitoring.hints.persistence.rules[].name — startup fails if the shortcut remains in a rule. Migration: move rule matching to where filters such as hint-type=CONFIG or severity=ERROR.

    monitoring.hints.persistence.rules = [
    { action = exclude, name = "CONFIG" }
    { action = exclude, where = "hint-type=CONFIG" }
    ]

Added

  • Observe shows where query time goes, and how it trends — Queries gains an Insights tab that separates planning, query-capacity wait, database-connection time, and time to the first result, and charts short-term history for the signals behind a slowdown: query concurrency and rejections, which kind of search is getting slower, ingest lag, throughput and queue pressure, service restarts, and memory growth. The ingest and system views gain matching history panels, and the golden-signal strips become sparklines. A Query Diagnostic Report shows the same breakdown for one captured request, and redacted support exports keep these value-free timings, so performance problems can be investigated without exposing query values. History covers roughly the last two hours at three zoom levels — recent minutes in detail, older data coarser — and lives only in memory: after a restart the charts show an explicit gap and start fresh instead of reporting zeros, as do captures taken before this release. It is on by default and costs well under a megabyte; disable or bound it under monitoring.stats-history.

  • Performance insight report for support — When you report a performance problem, one download now carries the evidence: the performance insight report (Observe → Queries → Insights) packages recent metrics history, query stage timings, the slowest search patterns, ingest health, and database pressure indicators into a single redacted JSON file safe to attach to a support ticket. You need no Prometheus or Grafana of your own. Search identifiers stay stable and usable after redaction, each disclosure tier enforces its documented size limit, and large downloads use gzip when the client accepts it. A full variant for local investigation is available at /observe/performance-insight/export/full. As with the other reports, the email action opens a pre-addressed draft for you to attach the download to; the report itself is never embedded in the email.

  • Repeated searches can use a faster database planning choice — IdentityScribe measures alternatives against live traffic and applies a faster choice to search patterns whose evidence is current, keeping the standard choice active by default. Once the evidence is stable it promotes the choice without an operator enabling it per pattern, and it keeps rechecking, so a choice returns to the standard one when its evidence weakens or changes. Decisions are never stored or keyed by the values you searched for. Observe shows the effective choice, the evidence behind it, and whether a decision is still learning, settled, or being reconsidered; the global learning control remains available for emergency rollback.

  • Observe shows which index build failed and why — When a background index build fails, the Observe → Storage → Build Queue view now lists each failed build by name with the reason it failed, instead of only a failure count. When there are more failures than the list can show, or the details have not loaded yet, the view says so alongside the count. If the same build returns to pending after a completed or failed attempt, the new lifecycle stays visible. Operators can see what needs attention without reading the server logs.

  • Incomplete-notification Inbox cards name the attributes dropped on repair — When the directory sends change notifications without objectClass, the Inbox card for repaired notifications now lists the attribute names recently dropped (for example objectClass, mail) instead of pointing operators only to ingest logs. It keeps a bounded set of the newest samples per entry type, so one noisy type cannot crowd out the others, and a recording failure never blocks the repaired entry. Samples contain attribute names only, never entry values.

  • Browser responses carry baseline security protections — IdentityScribe’s HTTP channels and published documentation now send headers that prevent framing, MIME-type guessing, and referrer disclosure.

Changed

  • The embedded REST API reference is updated — The /api and /observe reference pages now use the current stable release, including fixes for schema display and keeping request examples in sync.

  • Telemetry labels now use bounded, documented values — MCP tool counters and timings use fixed endpoint, tool, and result values; ingest failures use the stable reasons process, persist, and executor; and request-provided attribute names outside the configured catalog no longer create new metric label values. Operator dashboards can group these signals without labels growing from unexpected traffic values.

Fixed

  • Sorted searches with missing sort values are faster and remain complete during concurrent updates — IdentityScribe now chooses the fastest verified complete execution for common first-page searches automatically. Fast searches no longer spend milliseconds on planning work that can exceed the search itself, and entries without the requested sort value cannot be omitted because the data changed between planning and execution. Results, ordering, and cursors are unchanged; no configuration is required.

  • Operator UI presets survive immediate reload or tab close — Saving, updating, or removing a preset previously waited briefly before writing to browser storage, so reloading or closing the page right afterwards could silently discard a change that already appeared saved. Preset changes are now stored immediately.

  • Telemetry shutdown problems are now visible — A failed or timed-out flush of buffered telemetry during shutdown previously looked identical to a clean close. IdentityScribe now logs a warning naming the affected provider and step, so operators can detect telemetry loss at shutdown, and always continues closing the remaining providers.

  • Sizes below one byte no longer render an undefined unit — Fractional byte values in the operator UI displayed output such as 512 undefined; they now always render with a valid unit.

  • Large sorted pages no longer fall back too early after finding matching entries — In very large directories, a cursor-paged sorted search with a long list of possible attribute values could switch to slower behavior even while it was still returning matches. A search that is producing results now keeps going for longer before falling back, so the same page returns much faster, with results and ordering unchanged. To restore the previous behavior, set database.continuation.budget-extension-enabled=false; database.continuation.window.absolute-scan-cap remains the hard ceiling.

  • Synchronization stops safely after a transcription worker fails — A worker failure could leave accepted entries waiting forever during drain or allow queued work to begin after the failure. IdentityScribe now wakes drain and shutdown waiters, accounts for accepted work that cannot run, refuses later processor admissions, and keeps checkpoint advancement disabled so a restart cannot skip abandoned work.

  • Cursor-paged searches keep requested virtual attributes — A later page that selected a virtual attribute could omit it from the results. Such requests now preserve every requested virtual field on every page.

  • Range filters keep the exact bounds you send — Multiple range conditions on the same attribute could be narrowed through case-insensitive value comparisons before execution, changing the intended result for mixed-case or locale-sensitive values. IdentityScribe now preserves the original bounds and lets the configured directory comparison rules decide the match.

  • Range filters match entries whose bounds are met by different values — Lower and upper bounds on a multi-valued attribute are now evaluated independently, as required by LDAP filter semantics. Entries are no longer omitted when one value satisfies the lower bound and another value satisfies the upper bound.

  • Interrupted database write confirmation is no longer replayed blindly — If the database connection failed while confirming a completed entry update or delete, IdentityScribe could retry work whose outcome was unknown, risking duplicate history or a repeated delete. It now stops the affected worker and keeps its checkpoint pinned until the outcome can be resolved. Failed write preparation also releases partially opened resources.

  • Browser session cookies reject unexpected encryption choices — Session-cookie decryption now accepts only the key-wrapping and authenticated-encryption algorithms IdentityScribe itself uses. Existing valid sessions keep working; cookies that advertise a different algorithm fail authentication before their contents are processed.

  • Cancelled MCP change-history requests stay cancelled — A client cancellation during history streaming was wrapped as a generic server failure. MCP clients now receive the cancellation result, while real history failures keep their existing error handling.

  • Disabled database time limits remain disabled after temporary request settings — A connection that began with no query or lock time limit could retain a temporary limit after a request-specific override. The original disabled state is now restored, preventing unrelated later requests from failing early.

  • Machine-readable tokens stay stable under non-English host locales — Configuration values, HTTP preference handling, GraphQL change types, telemetry labels, export labels, and MCP reference lookups now use locale-independent casing. Hosts using Turkish or another locale with special casing rules produce the same protocol values and accept the same documented inputs as other hosts.

  • Observe’s API reference covers every shipped diagnostic endpoint — The generated Observe OpenAPI document now includes memory snapshots and exports, heap-dump capture outcomes, query diagnostic reports, the setup support report, and the history and performance insight downloads, with their real access levels, disclosure tiers, status codes, and response schemas.

  • Startup fails with a clear error instead of hanging when a service never finishes starting — Services that start in parallel during process startup now share the same five-minute ceiling that sequential startup already enforced. A service stuck in its startup phase previously left the whole process waiting forever without becoming ready or exiting; startup now stops the other services and fails with an error naming the stuck service, so process managers can restart or alert instead of waiting on a process that will never come up.

  • Shutdown during startup no longer briefly starts further services — When a shutdown request arrives in the short window while startup is still bringing service groups up, a group could slip past the cancellation broadcast, start its services after shutdown had already completed, and only then be torn down. Startup now re-checks for a pending shutdown at the moment each group is registered, so no service starts once shutdown has begun.

  • A failed checkpoint read no longer looks like a first-time sync — When reading the stored sync checkpoint failed after retries (for example the database was briefly unreachable during startup), IdentityScribe treated the result as “no checkpoint exists” and started a fresh full synchronization, discarding the saved position. A failed read now surfaces as an error and stops startup of the affected sync worker instead, so the durable checkpoint is preserved and the worker resumes from it once the database is reachable again.

  • The Getting Started curl download examples fail clearly on a bad download — The curl install snippets in Getting Started now use the same fail-fast, redirect-following flags as the Dockerfile example (curl -fSL --show-error). An unauthorized or failed download — for example an expired token — stops with a visible error at download time instead of saving an error response that only fails later at unzip.

  • Exported trace names can no longer carry directory or request data — Trace redaction covered the attributes attached to a trace but not the trace’s own name. A request whose URL matched no known endpoint previously used its raw request path — which can contain entry names — as the trace name; such requests are now named by their HTTP method alone. Any trace name outside the expected naming scheme leaves IdentityScribe as span.redacted instead of its literal text. Attributes remain governed by the existing redaction policy; no configuration change is needed.

  • Two transcribe types can no longer silently share the same storage — Transcribe type names that differ only in non-alphanumeric characters (for example user-domain and user.domain) resolve to the same storage, which would mix both types’ entries. Startup now rejects such a configuration with an error naming both colliding types, so the conflict is caught before any data is written. Existing configurations with distinct type names are unaffected.

  • Root DSE discovery requesting operational attributes is answered by IdentityScribe — An LDAP Root DSE search (empty base DN, base scope) that requests the all-operational-attributes projection + — the form many LDAP clients and browsers use for capability discovery — was forwarded to the connected directory, so clients saw the upstream directory’s capabilities instead of IdentityScribe’s. It is now answered locally like every other Root DSE search. Tree-browsing requests on real entries keep their existing forwarding behavior.

  • GraphiQL enforces the configured script integrity value — The channels.graphql.ui.asset.js-sri value was accepted but never reached the browser, so a tampered script from a custom CDN would still load. The GraphiQL page now carries that integrity value where the browser verifies it before executing the script. Browsers that do not support this check keep the previous behavior; the stylesheet integrity check is unchanged.

  • Opening GraphiQL while signed out redirects to the login page — With login enabled and anonymous API access disabled, opening /graphql in a browser returned a raw JSON authentication error instead of redirecting to the login page. The browser now lands on the login page and returns to GraphiQL after signing in. Schema and API requests keep the JSON authentication error.

  • /readyz reports not-ready when core services are unhealthy — The readiness probe only reflected synchronization readiness, so an instance whose core services (for example the database connection or a channel listener) turned unhealthy after startup kept answering /readyz with ok and stayed in load-balancer rotation. /readyz now also requires the services to be healthy, matching its documented “can safely receive traffic” contract. /livez, /startedz, and /healthz are unchanged, and sync or index backlogs still do not pull an instance out of rotation.

  • Searches naming several entry types with an OR filter are answered — A search over a base that spans several entry types and a filter listing the wanted types as alternatives — for example (|(type=person)(type=group)), alone or combined with further conditions — was rejected as if no type had been named at all. Such filters now narrow the search to exactly the listed types, the same way type=person already did. Filters that mix a type alternative with conditions on other attributes keep the previous behavior, because narrowing those could drop matching entries.

  • Breadcrumb links work for entries with long names — In the Operator UI, the breadcrumb trail on an entry page cut the entry identifier at 100 characters when building its link, so clicking a breadcrumb for an entry with a long distinguished name led to a page that does not exist. The link now always carries the full identifier; only the displayed text stays shortened.

  • Observe history panels preserve the evidence that still works — A malformed metric series no longer drops healthy series from the same collection. Query, ingest, JVM, service, and database views distinguish loading, first-load errors, stale data, and unavailable in-memory history; a later refresh or tier-switch failure keeps the last good charts visible with a retry action and stale-data notice.

  • Observe Inbox counts, lists, and details stay in sync — The sidebar count, Inbox list, and recommendation detail now share one bounded-freshness snapshot. Concurrent refreshes share one generation, and applying, dismissing, snoozing, or reopening a recommendation invalidates that snapshot before the next read, preventing a badge count from describing different evidence than the page.

  • Observe history and report downloads use the documented access checks — The in-memory history API now applies the same Observe read access as the other Observe JSON APIs, and the redacted and full report variants each require their documented access level.

  • Performance insight reports include incomplete-notification totals — The support report now includes recent skipped and repaired incomplete-notification counts when history is available. The Inbox card for skipped notifications also stays focused on the held-back updates, rather than listing dropped attributes, which apply only to repaired ones.

  • Heap dump capture no longer reveals server filenames — When on-demand heap dumps are enabled, the capture response now returns an opaque capture ID and size rather than the server location. Repeated capture attempts are also rate-limited with a retry time, so operators get a clear response instead of starting expensive captures back to back.

  • Monitoring configuration mistakes fail validation clearly — Telemetry log filters now reject conflicting operation-name constraints at startup, and one-sided saturation-pressure overrides are checked against the effective default for the other side. Invalid configurations fail with the normal monitoring configuration error instead of surfacing a lower-level startup exception.

  • Exact telemetry log filters stay exact — Monitoring log filters that name one operation exactly no longer also match near-named operations. Wildcard and regular-expression operation filters keep their existing pattern behavior.

  • Observe keeps configured attribute names in index-build labels — Storage build status now preserves the original configured attribute descriptor when reporting planned index work, so mixed-case attributes remain recognizable in the Operator UI instead of being shown through a normalized fallback.

  • Support downloads use safe filenames — Browser-triggered report downloads now sanitize unusual suggested names before setting the saved filename, preventing directory separators or control characters from appearing in download names.