v1.3.1
Release date: November 13, 2026
GreptimeDB v1.3.1 fixes write durability, database authorization, credential exposure, and query correctness issues. It also bundles Dashboard v0.14.6.
👍 Highlights
Write durability
Fixes potential data loss during repartition rollback or datanode restart, and restores periodic local WAL sync with the default sync_write = false setting. (#10253, #9715)
Database authorization and credential protection
Database selection through HTTP, MySQL, and PostgreSQL checks access permissions, and administrative functions require ADMIN privileges even through SELECT. Diagnostic configuration and startup logs redact credentials. (#10378, #10382, #10120, #10115, #10381)
Prepared query correctness
Fixes incorrect results from MySQL and PostgreSQL prepared statements after partition columns change, including incorrect rows written by prepared INSERT ... SELECT statements. (#10244)
Dashboard v0.14.6
Table Query and Metrics Query paginate their sidebars, and table pickers search by name. The SQL builder preserves the selected database, while log and trace views retain column-display preferences. (Dashboard #673, Dashboard #674)
Upgrade and compatibility notes
- Prometheus Remote Read: Tables with non-string label columns now return an error; timestamp and sample-value columns are exempt. Label matching follows Prometheus semantics: regular expressions match the full string, and missing or NULL labels match as empty strings. (#9633)
- Database selection: MySQL connections without a usable default database must select an authorized database before running statements that require one (error 1046). PostgreSQL
SET search_pathselects the first authorized, existing database and clears the selection if none qualifies. (#10382) - JSON log ingestion: Malformed escapes, out-of-range numbers, and nesting deeper than 1,024 levels are rejected under the configured
ignore_errorspolicy. Invalid UTF-8 remains recoverable. (#10245, #10302) - OTLP log tables: Direct OTLP ingestion auto-creates tables with the
greptime_log_v1schema; use a new destination table to adopt it, as existing tables retain their schema. During a rolling upgrade, avoid concurrent creation of the same destination table by old and new frontends; pipeline-defined tables are unchanged. (#10207) - Distributed deployments: Keep datanode clocks synchronized and stop the old process before reusing its node ID; process fencing takes full effect after all metasrv nodes are upgraded. A write whose leader lease expires during WAL append can return a retryable error after being applied, so retries can duplicate rows in append-only tables. (#9878, #9984, #9904)
- Manually configured sparse encoding: Before upgrading, check tables with
primary_key_encoding = 'sparse'or the legacymemtable.partition_tree.primary_key_encodingsetting against the requirements in #10020; unsupported schemas can fail validation and are not migrated automatically. Tables created normally by the metric engine and ordinary dense tables are unaffected.