Christian Schwarz 74df4a7b76 fix(walredo): walredo process that causes errors is never killed ...

Before this PR, if walredo failed, we would still update `last_redo_at`.
This means the `maybe_quiesce()` would never kill that process, although
clearly something is wrong.

However, we don't want to kill the process immediately on each redo failure,
because, the redo failure could be caused by corrupted or malicious
data.

So, change `maybe_quiesce()` to determine inactivity based on last
*successful* `last_redo_at`.
The result is that a transient redo failure won't cause a kill.
If there are *only* redo failures, we'll spawn & kill walredo process
at frequency 1/(10*compaction_period).

2024-02-01 14:38:53 +01:00

..

benches

pageserver: make walredo shard-aware

2024-01-03 14:22:40 +00:00

client

control_plane/attachment_service: complete APIs (#6394)

2024-01-31 12:23:06 +00:00

ctl

integrate tokio-epoll-uring as alternative VirtualFile IO engine (#5824)

2024-01-26 09:25:07 +01:00

pagebench

fix(pagebench): avoid CopyFail error in success case (#6443)

2024-01-23 15:11:32 +01:00

src

fix(walredo): walredo process that causes errors is never killed

2024-02-01 14:38:53 +01:00

test_data

Add walingest test (#5892)

2023-11-28 12:50:53 -05:00

Cargo.toml

integrate tokio-epoll-uring as alternative VirtualFile IO engine (#5824)

2024-01-26 09:25:07 +01:00