What rqlite/rqlite shipped

This week: SQL backups keep rows from quoted identifiers, GCS and S3 auto-backups support encryption keys, load timings return correct durations, status fields keep integer precision, and SQLite is now 3.54.1.

This week in rqlite: SQL dumps now preserve data from tables with quoted identifiers, GCS auto-backups can use customer-managed Cloud KMS keys, S3 auto-backups support server-side encryption, and load timings no longer return a negative duration. Status field selection keeps integer precision, and node bootstrap and data directory checks no longer change voter status or create Raft state unexpectedly. Code review of the db, store, internal and auth packages fixed several defects, and SQLite was upgraded to 3.54.1.

V10.6.6 is out. Node management endpoints remove, promote and demote now return JSON error bodies, local auto-backup rejects reserved filenames, backup files are kept if metadata rename fails, and a new test checks CDC events survive a Raft snapshot restore.

Rqlite v10.6.6 is out this week. The remove, promote and demote node-management endpoints now return JSON error response bodies instead of plain text. Local auto-backup rejects filenames reserved for internal use, so a configured name can no longer be overwritten by the client's own metadata. A backup file is now kept if the metadata rename fails. We also added a test showing CDC events survive a Raft snapshot restore, which previously had no store-level coverage. Dependencies were upgraded via go get.

Rqlite v10.6.5 and v10.6.4 are out. A rejected inter-node backup request can no longer stream the database, and authorization is now a single gate in the cluster service. GET /leader and POST /leader return JSON errors, and node probes honor request context.

Rqlite v10.6.4 and v10.6.5 are out this week. The most important change is security-related: the inter-node backup stream handler previously wrote an unauthorized response and then streamed the whole database anyway. Authorization is now a gate in the cluster service, and connections are closed on failure, so no command handler can skip the check. The leader endpoints also return JSON error bodies, POST /leader returns 404 for an unknown target node, and the backup stream handler checks for a nil request directly. Node probing for /nodes and /leader now honors the request context, and the db package makes the connection role first class. If you run rqlite with authentication, upgrade.

Rqlite v10.6.0, v10.6.1 and v10.6.2 are out. CDC events now carry a commit timestamp, the CDC HTTP sink follows redirects, and non-finite floats no longer skip a batch. The console schema view handles reserved JavaScript property names, and.backup and.dump truncate existing outpu

Rqlite v10.6.0, v10.6.1 and v10.6.2 are now available. This week's changes include a commit timestamp on CDC events and correct redirect following in the CDC HTTP sink. Non-finite float values are now encoded rather than causing a whole batch to be skipped. The console schema view now handles table names that collide with reserved JavaScript properties. The CLI.backup and.dump commands truncate existing output files so old data cannot corrupt the result.

This week: promote non-voting nodes to voters, demote voters to non-voters, several CDC fixes for dropped and lagging batches, snapshot threshold checked every 5 seconds, and gzip memory optimized with sync.Pool.

This week in rqlite: non-voting nodes can be promoted to voters, and voting nodes can be demoted to non-voters while staying in the cluster as read-only replicas. Several CDC fixes landed, including advancing the high watermark past dropped batches, keeping a follower's high watermark from moving backwards, and always answering snapshot sync requests so snapshotting is not blocked. The Raft Snapshot threshold is now checked every 5 seconds by default, and gzip memory usage is optimized with sync.Pool. Bound TEXT and BLOB bytes now count toward the compression size threshold.

V10.5.2 is out. Better CDC event handling, improved autocheckpointing at connect time, and more resilient cluster behavior when nodes encounter SQLite errors.

V10.5.2 released with improvements to change data capture reliability and cluster stability. Node-local SQLite errors now trigger exit to prevent cluster divergence, CDC events emit for each bulk request, and autocheckpointing is disabled at connection creation. New Cluster UI tab and configurable queued-writes retries give operators better visibility and control.

Rqlite v10.3.6 is out. Backup operations now report bytes written and file size. Snapshot ordering is fixed when the system clock moves backward. Rename and Remove operations are retried for better reliability.

Rqlite v10.3.6 is available. This release improves backup operations by reporting bytes written and backup file size, fixes snapshot ordering when system clocks move backward, and adds retry logic to Rename and Remove operations for better fault tolerance in distributed environments.

Rqlite v10.3.5 is out. Configurable slow-query logging, snapshot filtering, stronger PRAGMA validation, and CDC event fixes improve observability and reliability.

Rqlite v10.3.5 released with configurable slow-query logging for SQL observability, snapshot filtering by term and index, reinforced PRAGMA syntax validation, and fixes for CDC event handling in failed transactions. These improvements enhance debugging, filtering, security, and distributed database reliability.