Arc 26.09.3

[xe-nvdk]

Arc v26.09.3

Arc is a high-performance time-series database built on DuckDB, optimized for IoT, observability, and analytics workloads.

What's Changed

• fix(filereplication): a full-size .part with no final file is not "present" by @pujitha24 in fix(filereplication): a full-size .part with no final file is not "present" #965
• feat(metrics): expose per-peer replication lag gauges by @efegokdemir in feat(metrics): expose per-peer replication lag gauges #908
• fix(backup): skip oversized backup destination keys by @efegokdemir in fix(backup): skip oversized backup destination keys #903
• fix(query): rewrite comma cross-join tables to storage paths (query: a comma cross-join FROM a, b rewrites only the first measurement — the second stays a bare name and DuckDB fails with "Table ... does not exist" #978) by @xe-nvdk in fix(query): rewrite comma cross-join tables to storage paths (#978) #980
• feat(backup): name skipped files, carry the counts into the listing, add arc_backup_skipped_files (backup: a file skipped for an overlong destination key is a count only — the manifest names no key, the listing summary hides the skip, and no gauge reports it #977) by @xe-nvdk in feat(backup): name skipped files, carry the counts into the listing, add arc_backup_skipped_files (#977) #981
• fix(rbac): make permission extraction and query rewriting agree by @xe-nvdk in fix(rbac): make permission extraction and query rewriting agree #986
• fix(query): arm the replacement-scan guard on every relation keyword by @xe-nvdk in fix(query): arm the replacement-scan guard on every relation keyword #992
• fix(rbac): make RBAC enforcement real, and gate the listing endpoints by @xe-nvdk in fix(rbac): make RBAC enforcement real, and gate the listing endpoints #1000
• fix(delete): validate the WHERE fragment for table-position literals by @xe-nvdk in fix(delete): validate the WHERE fragment for table-position literals #1001
• fix(cq): re-validate the stored definition before each execution by @xe-nvdk in fix(cq): re-validate the stored definition before each execution #1002
• fix(filereplication): try the next peer on a checksum mismatch, keep the local copy (Checksum mismatch stops the peer loop, so one stale origin can block a path permanently #999) by @xe-nvdk in fix(filereplication): try the next peer on a checksum mismatch, keep the local copy (#999) #1003
• Reject unsafe edge-sync names by @efegokdemir in Reject unsafe edge-sync names #996
• Validate continuous query database names by @efegokdemir in Validate continuous query database names #995
• fix(cq): re-validate the stored database name before each run (CQ execution does not re-validate the stored database name, so a legacy row writes into the reserved storage root #1010) by @xe-nvdk in fix(cq): re-validate the stored database name before each run (#1010) #1013
• fix(ingest): start flush timeouts at dequeue and stop Close abandoning flushes (ingest: flush task timeout starts at enqueue, so queued tasks expire before a worker reaches them #1006, ingest: Close cancels in-flight flushes and discards queued ones, losing acknowledged writes on every graceful stop #1007) by @xe-nvdk in fix(ingest): start flush timeouts at dequeue and stop Close abandoning flushes (#1006, #1007) #1017
• fix(sql): disable DuckDB Hive partition inference on every read (read_parquet never sets hive_partitioning=false, so a key=value directory silently overwrites a real column #1005) by @xe-nvdk in fix(sql): disable DuckDB Hive partition inference on every read (#1005) #1018
• fix(ingest): defer flush when queue is full by @efegokdemir in fix(ingest): defer flush when queue is full #997
• fix(compaction): count rows from parquet file metadata by @efegokdemir in fix(compaction): count rows from parquet file metadata #1019
• docs: credit fix(ingest): defer flush when queue is full #997 in contributors by @xe-nvdk in docs: credit #997 in contributors #1024
• fix(cq): validate continuous-query measurement names, at the boundary and before each run (CQ source_measurement has no name rule, and update can blank destination_measurement #1011) by @xe-nvdk in fix(cq): validate continuous-query measurement names, at the boundary and before each run (#1011) #1022
• fix(ingest): re-enqueue a deferred flush when a worker frees a slot (ingest: nothing re-enqueues a deferred buffer, so flush workers idle while records wait #1008) by @xe-nvdk in fix(ingest): re-enqueue a deferred flush when a worker frees a slot (#1008) #1027
• docs(contributing): explain the arcx label and the unbuildable arcx_engine tag by @xe-nvdk in docs(contributing): explain the arcx label and the unbuildable arcx_engine tag #1029
• fix(build): run cargo from the arcx directory so its cargo config applies by @xe-nvdk in fix(build): run cargo from the arcx directory so its cargo config applies #1028
• fix(config): stop overriding DuckDB's own container-aware memory and thread limits (config: DuckDB memory_limit is derived from NumCPU x 2GB, so a quota-limited container gets a limit many times its own #1026) by @xe-nvdk in fix(config): stop overriding DuckDB's own container-aware memory and thread limits (#1026) #1031
• fix(api): mark panicked arcx Arrow streams as truncated by @efegokdemir in fix(api): mark panicked arcx Arrow streams as truncated #904
• fix(api): register arcx Arrow IPC queries in query management by @efegokdemir in fix(api): register arcx Arrow IPC queries in query management #911
• test(api): decode msgpack with the Basekick fork, not upstream v5 by @xe-nvdk in test(api): decode msgpack with the Basekick fork, not upstream v5 #1033
• fix(config): derive compaction.threads from the container's CPU quota (config: three NumCPU-derived defaults size for the host, not the container's CPU quota #1030) by @xe-nvdk in fix(config): derive compaction.threads from the container's CPU quota (#1030) #1035
• fix(wal): checkpoint flushed entries before recovery by @efegokdemir in fix(wal): checkpoint flushed entries before recovery #998
• docs: release notes and credit for WAL flush checkpoints (wal: replay after a crash re-ingests already-flushed records — permanent duplicates for tagless measurements #948) by @xe-nvdk in docs: release notes and credit for WAL flush checkpoints (#948) #1044
• fix(ingest): checkpoint WAL entries flushed on the asynchronous path (wal: asynchronous flushes record no checkpoint, so #948's duplicate replay is still live on the main ingest path #1045) by @xe-nvdk in fix(ingest): checkpoint WAL entries flushed on the asynchronous path (#1045) #1046
• feat(arcxrouter): HAVING and the bucket alias join the grouped allow-list by @xe-nvdk in feat(arcxrouter): HAVING and the bucket alias join the grouped allow-list #1047
• fix(wal): make replay idempotent by inheriting the replayed entry's identity (wal: purge by sequence floor instead of mtime, and delete a replayed file only after its data is flushed #1009) by @xe-nvdk in fix(wal): make replay idempotent by inheriting the replayed entry's identity (#1009) #1048
• feat(wal): per-file sequence bounds, a flush-aware purge, and an ordered Rotate (wal: purge by sequence floor instead of mtime, and delete a replayed file only after its data is flushed #1009) by @xe-nvdk in feat(wal): per-file sequence bounds, a flush-aware purge, and an ordered Rotate (#1009) #1049
• fix(arcxrouter): shadow compared only the FIRST batch, so it alarmed on results arcx got right by @xe-nvdk in fix(arcxrouter): shadow compared only the FIRST batch, so it alarmed on results arcx got right #1050
• fix(tiering): keep a cluster node's tier metadata in step with its own disk by @xe-nvdk in fix(tiering): keep a cluster node's tier metadata in step with its own disk #1062
• fix(wal): purge only data proven flushed by @lecodev-26 in fix(wal): purge only data proven flushed #1056
• fix(audit): copy Fiber request data before asynchronous logging by @0utsights in fix(audit): copy Fiber request data before asynchronous logging #871
• fix(cluster): keep Raft alive through the final flush so shutdown registers its files (cluster: Parquet files written during shutdown get no manifest entry, because the file registrar stops before the buffer #1014) by @xe-nvdk in fix(cluster): keep Raft alive through the final flush so shutdown registers its files (#1014) #1063
• fix(cluster): honor fetch cancellation during TCP/TLS connection setup by @pujitha24 in fix(cluster): honor fetch cancellation during TCP/TLS connection setup #902
• fix(storage): make ListStaged report only keys DeleteStaged accepts by @pujitha24 in fix(storage): make ListStaged report only keys DeleteStaged accepts #906
• fix(wal): line-protocol WAL entries carry their database envelope (cluster: line-protocol writes replicate into the default database, not the one they were written to #889) by @lecodev-26 in fix(wal): line-protocol WAL entries carry their database envelope (#889) #1055
• fix(cluster): ignore identical duplicate token creates by @efegokdemir in fix(cluster): ignore identical duplicate token creates #973
• fix(cluster): refresh file origin after failover rewrite by @efegokdemir in fix(cluster): refresh file origin after failover rewrite #984
• fix(cluster): make Raft manifest batches atomically visible by @efegokdemir in fix(cluster): make Raft manifest batches atomically visible #910
• fix(cluster): notify deletes during snapshot restore by @efegokdemir in fix(cluster): notify deletes during snapshot restore #983
• chore(release): prepare 26.09.3 by @xe-nvdk in chore(release): prepare 26.09.3 #1072

New Contributors

• @0utsights made their first contribution in fix(audit): copy Fiber request data before asynchronous logging #871

Full Changelog: v26.09.2...v26.09.3

Quick Start

Docker

docker run -d \
  -p 8000:8000 \
  -v arc-data:/app/data \
  ghcr.io/basekick-labs/arc:26.09.3

Debian/Ubuntu (amd64, arm64)

# Download and install
wget https://github.com/basekick-labs/arc/releases/download/v26.09.3/arc_26.09.3_amd64.deb
sudo dpkg -i arc_26.09.3_amd64.deb

# Start and enable
sudo systemctl enable arc
sudo systemctl start arc

# Check status
curl http://localhost:8000/health

RHEL/Fedora/Rocky (x86_64, aarch64)

# Download and install
wget https://github.com/basekick-labs/arc/releases/download/v26.09.3/arc-26.09.3-1.x86_64.rpm
sudo rpm -i arc-26.09.3-1.x86_64.rpm

# Start and enable
sudo systemctl enable arc
sudo systemctl start arc

Arch Linux / Omarchy (x86_64, aarch64)

# Download and install
wget https://github.com/basekick-labs/arc/releases/download/v26.09.3/arc-26.09.3-1-x86_64.pkg.tar.zst
sudo pacman -U arc-26.09.3-1-x86_64.pkg.tar.zst

# Start and enable
sudo systemctl enable arc
sudo systemctl start arc

Kubernetes (Helm)

helm install arc https://github.com/basekick-labs/arc/releases/download/v26.09.3/arc-26.09.3.tgz

Download Artifacts

Platform	Architecture	Package
Docker	amd64, arm64	ghcr.io/basekick-labs/arc:26.09.3
Debian/Ubuntu	amd64	arc_26.09.3_amd64.deb
Debian/Ubuntu	arm64	arc_26.09.3_arm64.deb
RHEL/Fedora	x86_64	arc-26.09.3-1.x86_64.rpm
RHEL/Fedora	aarch64	arc-26.09.3-1.aarch64.rpm
Arch Linux / Omarchy	x86_64	arc-26.09.3-1-x86_64.pkg.tar.zst
Arch Linux / Omarchy	aarch64	arc-26.09.3-1-aarch64.pkg.tar.zst
Kubernetes	-	arc-26.09.3.tgz (Helm)

Documentation

• Full Documentation
• Migration Guide

Community

• Discord: Join
• Issues: [GitHub](https://github.com/basekick-labs/arc/issues

---

This discussion was created from the release Arc 26.09.3.