Arseny Sher

Ignore zero sized cluster size in pageserver feedback.

1

To take into account that non zero shards send 0 size. Generally we should make reporting sharding aware, but for now this will avoid breaking size limit after https://github.com/neondatabase/neon/pull/6567

Skip deletion of safekeeper data if done previously.

1

To avoid re-listing s3 data during retries of tenant deletion with many branches.

We've seen in the wild one case where $subject. We had checked that there had been updates, but pageserver wasn't receiving them. Need to find the cause and fix it....

Stop postgres in immediate mode in VM image.

2

PR #6712 might add a delay in graceful shutdown without change to compute -> sk protocol to force sync of control file on safekeeper to persist commit_lsn, so use 'immediate'...

walproposer: move CHECK_FOR_INTERRUPTS to after waiting.

1

We should check it always after a sleep even before starting streaming.

Make WAL truncation atomic

1

See https://github.com/zenithdb/zenith/pull/847#discussion_r750131730

safekeeper: remove .partial suffix on the last WAL file.

2

Reasons: - it makes pg_waldump usage slightly more cumbersome, forcing to rename file. - it makes pull_timeline slightly more cumbersome because at any moment source file can be renamed from...

Epic: safekeepers dynamic membership change (Phase 1)

2

We need to be able to move tenants between safekeepers: 1. To migrate all tenants on one safekeeper if it dies. 2. To migrate one tenant if we find that...

Fix LR tests waiting for synced data.

1

Even if pg_stat_subscription.latest_end_lsn is caughtup, some tables might not be synced because until sync worker finishes main apply worker continues to advance. Context: https://neondb.slack.com/archives/C04DGM6SMTM/p1709307085456219 https://neon-github-public-dev.s3.amazonaws.com/reports/pr-6988/8112986047/index.html#suites/180444c850d4a41d41eb0a410dc16d84/1ff911895469b1af