Vlad Lazar 20788f0732 pageserver: allow read path to plan in-mem layer with concurrent write ...

Problem

Get page tracing revealed situations where planning an in-memory layer
is taking around 150ms. Upon investigation, the culprit is the inner
in-mem layer file lock. A batch being written holds the write lock
and a read being planned wants the read lock.

Solution

Lift the index out of the RW lock and allow the read path to operate on
an older version of it via ArcSwap. Note that the read IO operation
itself will wait for any on-going writes to finish, but at least we get
to plan concurrently with the write.

2025-05-08 14:22:25 +02:00

..

benches

feat: Direct IO for the pageserver write path (#11558)

2025-04-24 14:57:36 +00:00

client

storage_controller: coordinate imports across shards in the storage controller (#11345)

2025-04-24 11:26:06 +00:00

compaction

utils: explicit OTEL export config and OTEL enablement via common entry point (#11139)

2025-03-12 11:07:49 +00:00

ctl

pagectl: list layers for given key in decreasing LSN order (#11799)

2025-05-01 15:56:43 +00:00

pagebench

refactor: Extract some code in pagebench getpage command to function (#11563)

2025-04-19 08:38:03 +00:00

src

pageserver: allow read path to plan in-mem layer with concurrent write

2025-05-08 14:22:25 +02:00

test_data

pageserver: add layer visibility calculation (#8511)

2024-08-01 09:25:35 +00:00

Cargo.toml

feat: Direct IO for the pageserver write path (#11558)

2025-04-24 14:57:36 +00:00