LoginSign Up
← Back to all sparks
I

InfluxDB

ANALYTICS
Velocity5.0

Open-source time-series database for metrics and events

www.influxdata.com ↗

InfluxDB 3 is deep in a wave of data-correctness patching across three release lines as operators migrate to its Pacha Tree storage engine.

time-seriesstorage-enginedata-correctnesscompactionenterpriseobservability
◆Current state
InfluxDB 3 maintains three parallel release lines (v3.9.x, v3.10.x, v3.11.x) and a separate Enterprise tier, all publishing bug fixes in dense batches. The overwhelming focus of recent releases is data correctness in the storage layer: duplicate rows from concurrent snapshot handoffs, missing rows from snapshot persistence races, arbitrary overwrite resolution, and empty snapshot manifests creating sequence holes that stall compaction. The Pacha Tree storage engine upgrade is actively in use and generating its own class of operational issues — OOM on large source imports, index files left behind after retention, and stale run-set references.
◆Where it's heading
The product is converging its multi-line maintenance burden around storage engine migration correctness and compactor stability. Each line backports a common set of data-integrity fixes while Enterprise adds migration-specific features (retry command, startup phase logging, index backward compatibility). The privilege escalation fix in user authentication — present across 3.10 and 3.11 but currently off by default — signals that user auth is approaching GA. The trend is tighter data guarantees at the storage layer, not new capabilities.
◆Prediction
The next likely move is GA of the user authentication system currently in preview, alongside a continued push to close OOM and compaction edge cases as more deployments run the Pacha Tree storage engine upgrade at scale.

◆Recent moves

  1. 19d ago

    InfluxDB v3.11.2: Data-correctness fixes for snapshot races and WAL conflicts

    This release addresses six distinct data-correctness bugs in the Core storage layer, including missing rows during snapshot persistence (chunks evicted too early), duplicate rows from concurrent snapshot handoffs, and a partial schema mutation from a rejected line-protocol write. Enterprise adds a `retry-upgrade-to-pacha-tree` command that lets operators resume a stalled storage engine upgrade without data loss, plus compaction backlog and primacy livelock fixes for migration workloads.

    View source ↗
  2. 19d ago

    InfluxDB v3.11.3: Run-set index rollback safety, OR predicate file pruning fix

    Core has no changes in this release — v3.11.3 adds Enterprise-only fixes: the compactor can now read the v3 run-set index format written by later releases, enabling safe rollback from a newer version. A file pruning bug where OR predicates against an unindexed column silently dropped rows on the unindexed side is also resolved.

    View source ↗
  3. 19d ago

    InfluxDB v3.9.13: Snapshot sequence holes and WAL nonce fix backported to 3.9 LTS

    The 3.9 line picks up the snapshot manifest sequence-hole fix (empty snapshots now emit a manifest, preventing compaction stalls), the WAL nonce fix that prevented spurious node shutdowns on retried writes, and a panic fix for oversized duration values that could trap nodes in a startup crash loop. Enterprise adds startup phase logging and retention cleanup for orphaned file indexes.

    View source ↗
  4. 19d ago

    InfluxDB v3.10.6: Graceful shutdown timeout, catalog migration crash, privilege escalation fix

    v3.10.6 adds a configurable graceful shutdown drain timeout (`--shutdown-timeout`, default 30s) and fixes a catalog deserialization crash that blocked v2-to-v3 migrations when upgrading from Core. Enterprise closes a privilege escalation in the user authentication preview: non-admin subjects with `User:Update` could previously grant themselves permissions beyond their own role.

    View source ↗
  5. 19d ago

    InfluxDB v3.11.4: Write overwrite ordering fixed, OOM during storage engine upgrade addressed

    The overwrite resolution bug — where duplicate series key/timestamp pairs within a write buffer chunk were resolved arbitrarily due to an unstable sort — is fixed. Enterprise addresses two OOM scenarios during Pacha Tree storage engine upgrades (WAL replay and large source imports), each of which previously retried until the node ran out of memory. Plugin trigger cache invalidation is also corrected: disabling and re-enabling a trigger now re-reads the plugin source rather than resuming the stale cached copy.

    View source ↗