Skip to content

v0.5.9

Choose a tag to compare

@mbolli mbolli released this 17 Nov 06:45
· 5 commits to main since this release
  • backup: Config.ContentMD5 (<prefix>CONTENT_MD5) has Litestream send Content-MD5 on every upload that goes up in one piece, for a provider that requires it, as AWS does on an Object Lock bucket. An upload of more than 5 MiB goes up in parts, and Litestream sends no checksum on parts to a custom endpoint.
  • backup: lith's own S3 client, which aborts unfinished uploads and runs PurgeReplica's listings and deletes, uses Litestream's checksum policy: with an endpoint, checksums only where an operation requires one, and Content-MD5 on DeleteObjects.
  • backup: Manager.ReplicaNames(ctx) lists the replica names under Config.Prefix in the bucket or directory, names whose purge failed included, so an app can purge those its account list no longer has.
  • writer: Taken(ctx, q, key) reports whether a Once or OnceResult key has committed and is still kept, in a read transaction or a command's.
  • Docs: a command calls no other database's writer: neither Do nor Send makes the two commits atomic, Do makes the command's batch wait for the other writer's, and two databases that Do into each other deadlock. A follow-up in another database goes through a row the other side picks up. The skill, the review checklist and the commands guide say so.
  • backup: the docs cover versioned and Object Lock buckets (tested on Infomaniak's native S3, s3.gva2.infomaniak.cloud): Litestream needs nothing for them, and an erased account's data stays in old versions until they expire. Config.PurgeSettle's 15 s are for S3 over Swift (s3.pub2); on a provider whose listings are complete at once, a millisecond will do.