ArcadeDB 26.11.1

Overview

887 issues and pull requests closed under the 26.11.1 milestone - 651 issues and 236 pull requests - out of 483 pull requests merged and about 690 commits since 26.10.2. 468 of the closed issues are bug fixes, 28 of them rated critical.

The headline work is in six places:

Transactions that behave like SQL - a statement that fails inside a transaction is now taken back on its own (statement-level savepoints) and the transaction goes on, UPDATE ... SET reads the row as it was before the statement, an explicit NULL is stored as NULL, and CREATE INDEX no longer misses records other sessions have not committed yet.
Crash safety and durability - recovery after a power loss (Jepsen LazyFS) repairs torn pages and quarantines what it cannot recover instead of serving it, vector and sparse vector indexes survive a crash right after their build, and a growing record can no longer overwrite its neighbour on the page.
Vector search - the Cypher 25 SEARCH clause, Neo4j's CREATE VECTOR INDEX / CREATE FULLTEXT INDEX DDL, a pre-filter for db.index.vector.queryNodes, PRODUCT and BINARY indexes that search on their compressed codes and re-rank, and a graph build 2.3x faster on large INT8 indexes.
Operations - databases nobody uses can close and reopen on demand, a database that is only read holds no WAL file, every Docker image runs as UID 1000, the native image states its CPU baseline (with a compatibility archive for older x86 CPUs) and is built at -O3, and Claude Code can connect to your own server through the new ArcadeDB plugin.
Wire protocols - the PostgreSQL catalog answers what psql, pgjdbc and psqlodbc ask (\d, \dt, \di, \dm, \df, \du, JDBC metadata), PostgreSQL results stream in chunks, Redis pipelining runs at 0.8-0.9x of Redis on one connection and 2x under heavy load, and every protocol re-resolves a database after a close, restore or drop.
Security hardening - the users, groups and API-token files are never replaced by a default on a bad reload, plugin routes answer only the method they declare, wrong Basic credentials answer 401, and the HA cluster token can no longer change silently on a rolling restart.

Upgrading is recommended for every deployment. This release changes several answers to be correct (SQL, Cypher temporals, string functions, HTTP status codes), changes the Docker user and drops the plain Kubernetes StatefulSet: read the Upgrade notes before upgrading.

Contents: New features · Highlights · Security · Performance · Major fixes and improvements · Upgrade notes · Dependency updates

New features

Vector and full-text search

Cypher 25 SEARCH clause (#10776, #10783). MATCH (d:Doc) SEARCH d IN (VECTOR INDEX idx FOR $q [WHERE p] LIMIT k) SCORE AS s runs the vector search first and binds d to the k nearest records; the inner WHERE is tested on each candidate during the search, which widens until it has k matching rows. It used to be parsed and dropped, answering every Doc.
Neo4j index DDL in openCypher (#10678, #10864, #10914, #10994, #10915, #10992). CREATE VECTOR INDEX and CREATE FULLTEXT INDEX build LSM_VECTOR and FULL_TEXT indexes, with IF NOT EXISTS, OPTIONS from a map or a parameter, several labels or relationship types (FOR (n:Article|Post)) under one logical name, and a vector index that takes its dimension from the first vector it receives. SHOW INDEXES lists a createStatement that runs as written. Requested by @chetto1983.
IS :: VECTOR<T>(n) property type constraint and a vector(v, n, T) that builds the array of its coordinate type (#10862, #10918, #10917, #10991).
Pre-filter for db.index.vector.queryNodes (#10685, #10808): an optional options map with where and filter, tested during the search, so a multi-tenant query no longer returns fewer rows than asked. vector.neighbors and vector.sparseNeighbors take the same where option, evaluated only on the candidates the search visits (#10169).
PRODUCT and BINARY indexes search on their codes (#10514, #10968, #10975). The walk scores on the in-memory PQ codes (or the one-bit codes) and re-ranks k * rerankOversample candidates on the stored vectors, so the returned distances stay exact. New index settings rerankOversample and fusedPQ, a per-query rerank option, and pqTrainingLimit is now honoured. A PRODUCT graph file is about 8x smaller.
Full-text filters decided before the load (#11069): a condition on another indexed property next to SEARCH_INDEX() / SEARCH_FIELDS() is checked against that index first; with LIMIT 10 the reported query loads 10 records instead of about 2,000. Reported by @shaunpatterson.

Server and operations

Close idle databases and reopen them on demand (#11068). Off by default: arcadedb.server.databaseIdleTimeout closes a database after that long without activity, arcadedb.server.maxOpenDatabases keeps at most that many open (least recently used first), and the next request over any protocol (HTTP, gRPC, PostgreSQL, Bolt, Gremlin, MCP) reopens it. A database with an open transaction, a running query or async work, or under HA, is never closed. Requested by @singh-hovr.
WAL files and housekeeping on demand (#10234, #10218). One shared scheduler replaces a timer thread per database, a WAL file is opened by the first commit that needs it and released after 60 s idle (arcadedb.txWalIdleTimeout), so a database that is only read holds no WAL file. A background warm-up after ONLINE cuts the first HTTP query from about 160 ms to 60 ms (arcadedb.server.queryWarmUp).
MCP for Claude Code (#11170). The repository is a Claude Code plugin marketplace (claude plugin marketplace add ArcadeData/arcadedb --sparse .claude-plugin claude-plugin, then claude plugin install arcadedb@arcadedb) to connect Claude Code to your own server with an API token. Every MCP tool now carries read-only / destructive / open-world annotations, and the query and execute_command tools refuse scripting languages.
Native image: built at -O3, about 25% faster on warm loops (#11070, contributed by @tae898); an explicit CPU baseline (x86-64-v3 on amd64, armv8-a on arm64) with a "CPU requirements" section in docs/native-image.md (#10581, #10715); and a new linux/amd64 -compat archive built with -march=compatibility for CPUs below x86-64-v3 (#10623, #11253).
arcadedb.fileSync (default true) skips every fsync when set to false, for test runs and throwaway databases; the unit-test lanes now run twice as fast (#11149).
arcadedb.txWalGroupCommit (opt-in): concurrent commits share one WAL flush; about 4x the commits per second for writers into the same type (#10499).
arcadedb.graph.ghostEdgeProbe: a database known to hold no dangling adjacency entries can skip the per-vertex probe of variable-length walks (#10658).
arcadedb.bucketSelfDescribingChunkLinks (opt-in, default false): the new chunk-link format that lets a multi-page record link position 0. A database that stored it cannot be read by an older version (#10853, #10949).

Query languages and engine

Time series samples can be deleted (#10342, #10428): DELETE FROM <ts type> WHERE <time range> [AND tag = value] and TRUNCATE TYPE remove samples from the mutable and sealed storage, replicate under HA and refresh the continuous aggregates over the type.
Graph algorithms: algo.personalizedPageRank takes direction and weightProperty, and algo.labelpropagation takes pre-assigned labels (seedProperty, fixedSeeds) and refuses unknown keys (#11146, #11145, both reported by @chetto1983).
Statement-level savepoints (#10346), see Highlights.
sum() and avg() accept durations, in SQL too (#10198).

Studio

The graph view keeps moving with live physics (node repulsion, edge springs, centre gravity), as Neo4j Browser does (#10705, contributed by @shaunpatterson).
Charts are drawn with Apache ECharts instead of ApexCharts, which stopped being open source; all six charts resize with their panel and follow the theme (#10579).
The Add Index dialog has a Build Mode option (SORTED) for LSM_TREE indexes (#10363); "Copy to clipboard" works on plain-http pages (#10239, reported by @pwitherton); the GraphML export carries vertex and edge properties (#11236).
A default crud server group fills the gap between admin and readonly (#10778, contributed by @gramian).

Highlights

Transactions: a failed statement is taken back, and SQL reads like SQL

A statement that fails inside a transaction is undone on its own, and the transaction goes on (#10346, #10194, #10379). A multi-row INSERT, UPDATE, DELETE, CREATE EDGE, openCypher or sqlscript statement that failed after writing part of its records used to leave them in the caller's transaction, and the commit published half a statement. The engine now rolls the statement back the way SQLite does: its records, unique, plain, vector and full-text index entries (with the BM25 counters), edges and both vertices' edge lists, page bytes, record counters and reserved slots. When it cannot be taken back exactly (above arcadedb.txStatementSavepointMaxEntries, 100,000 by default, or a statement that touched a Graph Analytical View, dropped a file, or ran Gremlin), the commit is refused with a TransactionException naming the reason.
UPDATE ... SET reads the row as it was before the statement (SET a = b, b = a swaps), and an explicit NULL is stored as NULL: a DEFAULT fills only a property that is left out (#10194).
CREATE INDEX no longer leaves out a record another session wrote and has not committed (#10588): it stayed invisible to every index-backed query and a UNIQUE index accepted a duplicate of its key. A transaction that began before the index now queues the entries it owes the index at commit. With it, CREATE INDEX in a transaction that wrote to the indexed type no longer loses every committed record on rollback (#10490), and a refused DROP INDEX no longer leaves the index half dropped (#10589).
A UNIQUE key moved off its holder can be given to only one record (#10265), a NOTUNIQUE_HASH key changed away and back no longer lists the record twice (#10269), and a compacted NOTUNIQUE LSM index no longer returns entries its records do not have (#10188).
A failed transaction() block that joined the caller's transaction leaves it open instead of rolling back the caller's earlier work (#10135); deleting a missing or vetoed record over the remote drivers now throws (#10230).
An update of a record with an undecodable property is refused instead of saving the record back without that property (#10955), and an auto-commit gRPC ExecuteCommand retries a transient conflict like HTTP does (#10942).

Crash safety, storage and HA

Power loss: crash recovery repairs torn pages whose earlier versions are still in the WAL, so a recovered node no longer returns records with wrong or missing property names (#11151, #11153), and a database whose recovery did not complete is quarantined (LOCAL_RECOVERY_INCOMPLETE, UNAVAILABLE_UNIQUE_INDEX) and resynced from the leader instead of serving half-recovered data (#11250). Found with the Jepsen LazyFS suite.
Vector indexes are durable at READY: a process killed right after CREATE INDEX or REBUILD INDEX of an LSM_VECTOR index no longer reopens it empty or loses the transactions committed after it (#11238). Committed LSM_SPARSE_VECTOR postings survive a crash through per-index posting logs, and every HA node answers with the postings the leader has not sealed yet (#11239, #11257). A full vector rebuild keeps the deletes and inserts committed while it reads the pages (#11099).
Records on a full page: growing a small record no longer writes its 9-byte pointer over the next record (#10617, #10637, #10669, #10859), a multi-page record can no longer be cut short at commit by a link to slot 0 of page 0 (#10701), and freeing a chunk chain no longer unlinks a live record (#10875).
Restores: an overwrite restore extracts into a staging directory and keeps the old copy on failure, so a mistyped encryption key no longer destroys the target (#11050, #11058, #11041); while a restore replaces a database it refuses every write on every protocol, so nothing is acknowledged and then lost (#10689, #10533); and under HA it waits for the leader commits still replicating and refuses replica writes (#10881, #10965). A snapshot install checks the installed files against the leader's account before the backup is deleted (#11200, #11223).
HA: an index compaction whose publication has an unknown outcome quarantines the leader instead of leaving its index silently short (#11240); a refused CREATE INDEX ... UNIQUE no longer quarantines the leader (#10506, #10569); eight apply-path failures no longer let the checkpoint move past an entry this node skipped (#10086); one Gremlin write sent to a follower is no longer applied several times (#11182); a replicated apply publishes its pages under the publication sequence protocol (#10941).
A database opened READ_ONLY writes nothing when it closes (#10507), and schema.prev.json is a byte copy, never a hard link, so it stays readable on SeaweedFS and similar file systems (#11097).

Graph and openCypher correctness

Concurrent edge deletes no longer leave an edge list naming a deleted edge or a reused RID (#10263, #10304), a lightweight-edge delete after reading an outdated list is no longer lost (#10302, #10315), a super-node delete removes exactly one twin (#10314), and an edge-list Iterator.remove() no longer loses its removal at commit (#10375).
A MutableVertex taken by modify() no longer saves a stale edge list over edges appended through another handle (#10929); DROP TYPE refuses an edge type whose edges are all lightweight (#10923, #11142); REBUILD TYPE ... POLYMORPHIC WITH repartition = true no longer rewrites subtype records as the scanned supertype (#11042, #11043, #11066).
CHECK DATABASE FIX on a super-node no longer deletes its lightweight edges (#10377), and deleting a vertex no longer leaves its self-loop edge records behind (#10520).
openCypher: an explicit write of a lookup key makes the query eager, so a MATCH filter sees the graph as it was before the write (#11173, #11230); temporal values compare, add and construct as in Neo4j (#10198, #10199, #10200); DISTINCT, GROUP BY, UNION, min() and max() key BINARY and ARRAY_OF_* values by content (#10380, #10195); dynamic labels in CREATE/MERGE are no longer dropped (#10683); and a long run of query-correctness reports from @YGY-001 is fixed (LIMIT 0 after WITH collect(), CALL { } IN TRANSACTIONS after an empty batch, WITH * and FOREACH visibility, among others: #10892, #10886, #10891, #10890, #10889, #10515, #10513, #10512, #11004, #11005, #11008).
Graph Analytical Views reach the commits in commit order (#10249), no longer count an edge twice after a background compaction (#11192), pin one snapshot per query (#10365), and need far less heap to build and restore: a 6M-edge view now builds in a 112 MB heap (#11011).

Time series

Deletes, TRUNCATE and late samples keep continuous aggregates right on every ingest path (SQL, Java API, async, line protocol, Prometheus remote write, gRPC, JSONL importer), under HA too (#10502, #10686, #10632, #10633, #10634, #10673).
WHERE ts > 9223372036854775807 no longer overflows to the whole range, where DELETE removed every sample (#10528); a type rename moves the engine's files (#10749); DROP TYPE serialises with a sealed-file merge (#10866, #10867); CREATE TIMESERIES TYPE refuses a sealed file already at its name (#10925).
Cypher, Gremlin and GraphQL refuse a time-series type instead of matching nothing (#10440, #10442), MongoDB QL reads it (#10468), and SQL INSERT refuses a number for a BOOLEAN or STRING column, as every other ingest path does (#10137).

Security

Authorization, input-bounds and hardening fixes found by audit and by reporters. The related GitHub security advisories are published with this release.

MongoDB wire: a legacy OP_QUERY on a collection no longer reaches documents without passing the authorization gate.
Per-type ACL: the first batch request to a closed database opened on demand now binds the per-type ACL; renaming a type with only updateSchema no longer escapes the ACL of the type; three schema listings, the cluster endpoint and the gRPC getDatabaseInfo filter what a user cannot read; full-text BM25 statistics respect bucket access control.
Security file reloads (#10321, #10326, #10383, #10403, #10451, #10460, #10404): a malformed, missing or unreadable server-users.jsonl, server-groups.json or API-token file on a running server keeps the users, groups and tokens in force instead of replacing them with a default (a passwordless root, the default groups), keeps the bad file as an -error copy, and a replicated change whose write failed is no longer undone by the watcher.
Read-only paths: on the native image, writes nested in a SELECT (and PROFILE INSERT) no longer pass the read-only query paths; analysing a Gremlin script no longer runs it, so the MCP read-only tool cannot commit a write (#11182); the MCP upsert tools honour allowSchemaChange, and MCP refusals no longer reveal which databases exist.
Plugin routes answer only the method they declare (#10157, #10296, #10385): GET /api/v1/cluster/stepdown no longer steps the leader down, and a plugin can no longer silently replace another plugin's route or a core route.
Authentication: wrong Basic credentials answer 401, not 403 (#10109); a failed Bolt LOGON no longer leaves the previous principal bound; the snapshot endpoint no longer falls back to Basic credentials on a wrong cluster token (#10827); a user's resultSetLimit and readTimeout apply to that user only (#11248).
HA cluster token: a node refuses to start when the root password behind its derived cluster token changed, from the file or inline (#10601, #10924), and a sender can no longer force a Raft channel reset or take a throttle slot by naming this node in an unauthenticated header (#10144, #10225, #10299).
Redis wire: an unauthenticated connection can no longer be held open at a trickle past the idle timeout, nor turn every junk byte into two SEVERE log records (40 KB of input wrote 6.2 MB of log) (#10597).
Operator intent: a /ws subscribe frame no longer reopens a database an operator closed (#10690).

Performance

Graph: a Cypher triangle count written as one MATCH takes 5 ms on a Graph Analytical View and about 40 ms without one, where it took about 1,750 ms (#10255); variable-length hops with distinct endpoints walk the view, 6 hops from 222 ms to 6 ms (#10548); algo.bfs stops at maxDepth (#10539); IN $list seeks the index with an OPTIONAL MATCH (#10354, contributed by @sainiteesh36).
Indexes and schema: CREATE INDEX ... UNIQUE on a populated type is linear instead of quadratic (#10334), the sorted build runs beside pending writes to other types (#10466), and schema changes scale with the types (#10247).
Vectors: the INT8/BINARY graph build keeps the quantized rows instead of churning a float cache, 2.3x faster at 1M x 768 with about 50x less GC pause time (#10543); ARRAY_OF_FLOATS is stored fixed-width, 9 rows of 1536 dimensions per page instead of 7 (#10856); a PRODUCT rebuild keeps its PQ codebook (#10847) and reopens in 0.7 s instead of 2.6 s on 1M vectors (#10568); hybrid queries load each candidate once (#10129).
Storage and allocation: a bucket insert no longer walks every slot of its page, 19% of the CPU of a small insert (#10576); LSM lookups share one page view, 36-41% less allocation (#10654); a filtered scan reuses its rejected rows (#10655); variable-length number reads allocate nothing (#10656); GraphBatch flushes one-bucket edge types in parallel (#10547). Several of these were contributed by @tae898.
Startup: the GraalVM polyglot engine is built only when a script language is used, so SQL and Cypher load no Truffle classes (#10245).
PostgreSQL: a SELECT of declared properties streams in chunks per fetch instead of being materialised whole (#6659, contributed by @GadiVenkatesh).
Redis: pipelined replies are sent together (0.35x to 0.8-0.9x of Redis on one connection, about 2x Redis with 200 connections), and in-memory commands bypass the query admission gate (#10499).

Major fixes and improvements

Wire protocols

PostgreSQL catalog: psql \dt, \d, \d+, \di, \dm, \df, \du and psqlodbc SQLColumns/SQLTables answer from the emulated catalog (#10228, #10473, #10691, #10591); pg_index, pg_depend, pg_matviews, pg_settings and the JDBC type lookups are modelled (#10473, #10495, #10842, #10961); a statement that merely contained pg_catalog in a literal is executed again (#10695); a connection re-resolves its database after a close, restore or drop (#10692); text-format array parameters keep empty and NULL elements (#10870); format_type() and regtype print PostgreSQL's names (#10228).
Redis: a key starting with $ is no longer the same key as the one without it (#10694), inline commands are answered (#10597), and a command whose name is not a string gets an error instead of silence (#10845).
MongoDB: every command uses the instance currently open after a restore or reopen (#10562), and writes on a time-series collection are refused unless applied exactly (#10464).
gRPC: graph projections and lists of a search hit arrive typed instead of as a Java toString or a JSON string (#11044, #11048, #11207), BeginTransaction refuses SERIALIZABLE (#10386), and authentication faults are logged and counted (grpc.auth.faults.total) (#10142).
A closed database is reported as unavailable, not as a server fault, on every wire protocol (PostgreSQL 57P03, Redis LOADING, MongoDB TemporarilyUnavailable, gRPC UNAVAILABLE, Bolt DatabaseUnavailable) (#10553, #10692).
HTTP: committing a session the server does not hold answers 404 instead of 204 (#10164); POST /api/v1/begin validates isolationLevel (#10295, #10337); the OpenAPI document works on the native image (#10573) and is accurate for the group listing (#11276).

SQL and functions

String methods count Unicode characters, not UTF-16 code units, in SQL and openCypher (#10248, #10378, #10523, #10524).
=, <>, IN and CONTAINS compare list and map elements as they compare scalars (#10595, #10448); DISTINCT, GROUP BY, min() and max() compare BINARY, array, map and temporal values the way = does (#10696, #10697, #10710).
A DATETIME string with an offset is stored as the UTC wall clock of the instant (#10489, #10596); a number into a DATE is a day count under every dateImplementation (#10910); invalid dates such as 2021-02-30 are refused (#10206).
On an edge, out and in read the endpoints (#10684); shortestPath() maxDepth counts edges (#10439), and dijkstra()/astar() bound the path (#10527); if, randomInt, pow() and the Cypher-style functions no longer truncate or wrap a number argument (#10699, #10850, #11047); positional and named parameters are counted, so an unbound placeholder is refused (#10721, #11098).
INSERT INTO T FROM SELECT ... FROM T no longer reads its own inserts (#10192), triggers fire for subtype records (#10207), REFRESH INCREMENTAL views follow writes to subtypes and joined types (#10205), and a refused schema statement changes nothing (#10317, #10412).

Packaging and deployment

Every Docker image, JVM, native and custom distribution, runs as 1000:1000 and declares the same five volumes (#10319, #10330, #10387); the compose quickstart persists all of them (#10402).
The Helm chart is the only supported Kubernetes deployment (#10542); a config directory moved off <root>/config, or a volume mounted over it, is seeded with the shipped defaults (#10536, #10554); an HA node started with only arcadedb.server.rootPasswordPath forms a cluster (#10462).
The native image creates GEOSPATIAL indexes (#10423) and ships a parseable users template (#10320).

Importers and remote clients

JSONL EXPORT/IMPORT DATABASE round-trips declared LIST/MAP OF DATE, BINARY, DOUBLE and record links (#10201, #10391, #10483, #10529); the JSON importer keeps integers and longs exact (#10756); GraphSON and GraphML export every property type (#10202).
The Java HTTP client reads +/-Infinity and NaN floats (#10899), sends a LocalDate parameter as a typed marker and reads LIST OF DATE items back as LocalDate (#10865, #10908, #10909, reported by @TobiasJoseHermann).
Studio's record editor no longer discards edits or retypes untouched properties on save (#10082, #10083).

Build and CI

CI runs Maven through ./mvnw, the slow and HA lanes run in parallel shards and the CI wall time dropped from 43 to 26 minutes (#10134, #11154, #11183, #11189, #10887, contributed by @agarg21); census tests enumerate from the authority they protect (#11199, #10982, contributed by @2748109647).

Upgrade notes

Each item links to the issue with the full migration detail; the complete write-up is in docs/release-26.11.1.md.

Docker images run as UID 1000 (#10319, #10330). A volume written by an earlier native image (root on amd64, 65532 on arm64) or custom-distribution image (100:101) must be handed over once with chown -R 1000:1000 <path>. Custom images now declare config and replication as volumes, and the compose quickstart mounts a named volume on all five: copy users, groups and tokens out of the old anonymous config volume before recreating the stack (#10387, #10402).
Kubernetes: config/arcadedb-statefulset.yaml is no longer shipped; move to the Helm chart at ArcadeData/arcadedb-helm (#10542).
HA cluster token: with a token derived from the root password (file or inline), the password can no longer be rotated with a rolling restart. Restart every node with arcadedb.ha.clusterTokenAllowRotation=true once, or use arcadedb.ha.clusterTokenPath (#10601, #10924).
HTTP: wrong Basic credentials answer 401 (was 403) (#10109); committing an unknown or foreign session answers 404 (was 204) (#10164); plugin routes answer 405 on an undeclared method, and registerAPI(HttpServer, PathHandler) is deprecated for removal (#10157, #10296); a plugin route that collides with another plugin's or a core route fails the server start (#10385).
Transactions: a failed statement inside a transaction is taken back and the commit publishes the rest (#10346); a failed joined transaction() leaves the caller's transaction open (#10135); UPDATE ... SET reads the pre-statement row and an explicit NULL is stored as NULL, so a NOTNULL property with a default refuses it (#10194).
SQL and openCypher answers that changed:
- string functions count characters, not UTF-16 code units (#10248, #10378);
- Cypher durations are not comparable with </>, a localdatetime() is not a datetime(), duration arithmetic is exact and calendar units follow the local calendar (#10198, #10199, #10200);
- a DATETIME string with an offset is stored as UTC (values already stored are not rewritten), and asDateTime(format) / date(value, format) follow (#10489, #10596);
- list and map elements compare as scalars, [1] = ['1'] is true (#10595); on an edge out/in are the endpoints (#10684); shortestPath() maxDepth counts edges, lower it by one if you raised it as a workaround (#10439);
- a fractional or out-of-range integer argument is refused by if, randomInt, vector.sparseCreate and the Cypher-style functions (#10699, #10850, #10954);
- a number into a DATE is a day count under every dateImplementation (#10910), an impossible date such as 2021-02-30 is refused, and a DATE compared with a DATETIME is midnight of that day (#10206);
- a SQL statement given more or fewer parameter values than it has placeholders is refused instead of dropping or binding null (#10721, #11098);
- Cypher, Gremlin and GraphQL fail on a time-series type instead of matching nothing (#10440, #10442), and SQL INSERT into a time series refuses a number for a BOOLEAN or STRING column (#10137).
Indexes and storage: a transaction that gives one UNIQUE key to two records now fails (#10265); an update of a record with an undecodable property is refused (#10955); a PRODUCT vector index ranks through PQ codes and re-ranks, raise rerank or rerankOversample if recall matters more than speed (#10514).
Remote drivers: deleting a missing or vetoed record throws RecordNotFoundException / RecordVetoedException (#10230); a LocalDate parameter is sent as a typed marker, so upgrade client and server together (#10908); gRPC BeginTransaction refuses SERIALIZABLE (#10386).
Wire protocols: PostgreSQL format_type() and regtype print integer, character varying (#10228); Redis closes a connection on a malformed message (#10597); MongoDB refuses updates and inexact deletes on a time-series collection (#10464).
HA: a compaction publication with an unknown outcome quarantines and resyncs the leader (#11240).

Dependency updates

About 75 dependency bumps landed in this cycle, almost all through Dependabot. The notable ones:

Server and engine: gRPC 1.84.0, GraalVM Polyglot 25.4, jvector 4.0.2 (native vectorization off unless configured), Jackson 2.22.4, Micrometer 1.17.2, OpenTelemetry 1.67.0, Logback 1.6.6, Gson 2.15.0, jboss-threads 3.11.1, lz4-java 1.12.1, snappy-java 1.1.10.12, JLine 4.4.8 and swagger 3.0.0. Guava is pinned to 33.7.2-android (CVE-2026-102554) (#11237).
Studio: Apache ECharts replaces ApexCharts, d3-force drives the live graph physics, cytoscape 3.34.4 and swagger-ui-dist 5.34.0; JSZip and the PDF export are gone, and the license check now fails the build on a package outside the allow-list (#10579, #10626).
Build and test: Surefire/Failsafe 3.6.1, JaCoCo 0.8.16, GraalVM native-maven-plugin 1.1.16, ASM 9.11.1, testcontainers, the Neo4j drivers of the e2e suites and the Go module bumps.
CI: GitHub Actions and pre-commit hook bumps.

To everyone who reported, reproduced, reviewed, tested and fixed - in particular @ruispereira, @tae898, @YGY-001, @chetto1983, @ItsDeadlyProgrammer, @shaunpatterson, @neomodular, @sainiteesh36, @YaeSakuraQ, @GadiVenkatesh, @agarg21, @gramian, @7487, @2748109647, @pwitherton, @TobiasJoseHermann, @FilBr, @singh-hovr, @shulei5831sl, @FEARIS2

ArcadeData/arcadedb 26.11.1 on GitHub