Commit Graph

3016 Commits

Author SHA1 Message Date
Christian Schwarz
f32ebb74a1 refactor: move the ldpreload setup code into a method on EvictionEnv 2023-03-29 15:55:13 +02:00
Christian Schwarz
c8784cba6b wire up the statvfs ldpreload thing in an example test 2023-03-29 15:44:14 +02:00
Christian Schwarz
555ccb8c91 feat: add LD_PRELOADable library for mocking statvfs
use like so:

env RUST_LOG=pageserver=info,pageserver::disk_usage_eviction_task=debug LD_PRELOAD=$PWD/target/debug/libstatvfs_ldpreload.so  NEON_STATVFS_LDPRELOAD_CONFIG="$(echo '{}' | jq '{magic: "foobar", mock: { type: "Failure", mocked_error: "EIO" }}')" ./target/debug/neon_local pageserver start
2023-03-29 15:01:45 +02:00
Christian Schwarz
216f613e24 Merge pull request #3890 from neondatabase/heikki/disk-usage-eviction
Rewrite parts of disk usage eviction implementation to make it more understandable (I hope).
2023-03-29 12:43:38 +02:00
Christian Schwarz
b47a02569f tests: fully read-only warmup + wait for remote storage upload 2023-03-29 11:55:02 +02:00
Christian Schwarz
699ca672a4 test: refine test_pageserver_respects_overridden_resident_size 2023-03-29 11:22:53 +02:00
Christian Schwarz
a698ddb8a4 fix: avoid needless timeline.clone() 2023-03-29 10:58:59 +02:00
Christian Schwarz
57d215e6bb fix: suggestions commited from GitHub web didn't compile 2023-03-29 10:55:57 +02:00
Christian Schwarz
83813f2cb1 fix: remove unneeded clippy allow 2023-03-29 10:55:45 +02:00
Christian Schwarz
bdc7f8d192 fix: remove now-unused is_sorted 2023-03-29 10:52:56 +02:00
Christian Schwarz
9a55e4f909 fix: structured logging of tenant_id
Co-authored-by: Joonas Koivunen <joonas@neon.tech>
2023-03-29 10:51:14 +02:00
Christian Schwarz
0b9a44a879 fix: structured logging of tenant_id
Co-authored-by: Joonas Koivunen <joonas@neon.tech>
2023-03-29 10:50:36 +02:00
Christian Schwarz
a6f9ebf178 fix: repeat tenant_id in debug message 2023-03-29 10:49:06 +02:00
Christian Schwarz
370b3637db doc: add explainer to debug_assert 2023-03-29 10:47:15 +02:00
Christian Schwarz
88753b3325 doc: link follow-up issue in TODO comment 2023-03-29 10:36:55 +02:00
Christian Schwarz
bb5947afde test: test_pageserver_respects_overridden_resident_size: use absolute wiggle room instead of percentage
Heikki added the `*0.75` in

    commit 11b16614a3
    Author: Heikki Linnakangas <heikki@neon.tech>
    Date:   Tue Mar 28 01:13:33 2023 +0300

        Fix test for change in behavior close to the min_resident_size boundary

        This PR changed the behavior to match my expectation per my comment:
        https://github.com/neondatabase/neon/pull/3809/files#r1149837135

Without it, the test fails because we fall back to global LRU, and we
have an assert on that.

The reason why it falls back to global LRU is that
`target = delta_between_small_and_big_tenant`
doesn't leave any wiggle-room to go over min_resident_size boundary.

But, we redefined min_resident_size to include up to 1 layer above it
in this branch.
Multiply that by two because we're dealing with 2 tenants here.
2023-03-28 19:16:59 +02:00
Christian Schwarz
d6c2867b46 doc: add debug_assert for self-documenting candidates.sort_unstable_by_kye() 2023-03-28 19:16:26 +02:00
Christian Schwarz
386c2d0112 refactor: go back to a single list
The MinResidentSizePartition is effectively what `overage` was earlier,
but more expressive and outside of EvictionCandidates.

So switch the code back to a single list,
but use (MinResidentSizePartition, EvictionCandidates) tuples.

That eliminates the need for iter_in_eviction_order() alltogether.

It consumes 8 bytes more memory per candidate, but, that doesn't matter
for now.
2023-03-28 18:25:44 +02:00
Christian Schwarz
704d4f4640 doc: improve comment on min_resident_size 2023-03-28 18:06:44 +02:00
Christian Schwarz
dc72a9534e doc: update doc comment for collect_eviction_candidates
And move the impl of MinResidentSizePartitionedCandidates
below it because it makes sense when reading the code top-down.
2023-03-28 18:05:31 +02:00
Christian Schwarz
07c44f9151 doc: hint that usage_assumed is modified in the loop 2023-03-28 17:47:11 +02:00
Christian Schwarz
0c10e6d3e7 feat: demote info logs to debug
These would be per tenant, we don't want to emit thousands of log lines
when this code runs.
2023-03-28 17:36:59 +02:00
Christian Schwarz
85becb148f feat: bring back min_resident=max(all layers) behavior 2023-03-28 17:36:59 +02:00
Christian Schwarz
ea3c76a9d6 refactor: instead of 'overage', have two separate lists 2023-03-28 17:36:59 +02:00
Christian Schwarz
799576ab1e Merge branch 'problame/disk-usage-eviction' into heikki/disk-usage-eviction 2023-03-28 15:24:52 +02:00
Joonas Koivunen
b1d54024e7 doc: why tokio mutex instead of std mutex 2023-03-28 14:14:14 +03:00
Christian Schwarz
18ed0f9a06 enable disk-usage-based eviction in staging
We'll do some manual testing of the statvfs loop there.
2023-03-28 12:52:16 +02:00
Christian Schwarz
453c3fd2da doc: fix typo
Co-authored-by: Heikki Linnakangas <heikki@neon.tech>
2023-03-28 12:22:26 +02:00
Christian Schwarz
0428b6822a doc: more comment on lru_candidates to address questions from review 2023-03-28 11:53:48 +02:00
Christian Schwarz
42d63270a5 doc: add comment on extend_lru_candidates 2023-03-28 11:44:10 +02:00
Christian Schwarz
0056108c45 doc: remove stray comment 2023-03-28 11:44:10 +02:00
Christian Schwarz
54cc1d5064 doc: sub-headings for mechanics & policy in module comment 2023-03-28 11:44:10 +02:00
Joonas Koivunen
70c837a4b2 refactor: simplify max_layer_size as u64 2023-03-28 12:44:01 +03:00
Christian Schwarz
38d3061143 add comment on not-need to be 100% accurate about max_layer_size 2023-03-28 12:44:01 +03:00
Joonas Koivunen
75759f709f doc: explain "bug" message, log layers 2023-03-28 12:44:01 +03:00
Joonas Koivunen
03ab5df081 chore: remove dead code 2023-03-28 12:44:01 +03:00
Joonas Koivunen
6e8d7b449f refactor: combine nested macthes 2023-03-28 12:44:01 +03:00
Joonas Koivunen
17b5c8d1c4 refactor: get rid of ApproxAccurate 2023-03-28 12:44:01 +03:00
Joonas Koivunen
0943dd30eb chore: clippy 2023-03-28 12:44:01 +03:00
Joonas Koivunen
b599755042 refactor: rename DiskUsageEvictionState => State 2023-03-28 12:44:01 +03:00
Joonas Koivunen
244185e6e6 doc: comment changes
Co-authored-by: Christian Schwarz <christian@neon.tech>
2023-03-28 12:44:01 +03:00
Joonas Koivunen
0a5043fae5 refactor: less static mutexes 2023-03-28 12:44:01 +03:00
Heikki Linnakangas
041b708dc6 rustfmt 2023-03-28 11:08:35 +03:00
Heikki Linnakangas
11b16614a3 Fix test for change in behavior close to the min_resident_size boundary
This PR changed the behavior to match my expectation per my comment:
https://github.com/neondatabase/neon/pull/3809/files#r1149837135
2023-03-28 01:28:27 +03:00
Heikki Linnakangas
b6b8265450 Rewrite to make the algorithm more understandable (I hope).
The algorithm is the same (with two small exceptions), but rewrite the
way it's implemented to make it easier to follow.

The exceptions:
1. 'min_resident_size' now protects at least that much data in the first
   "respectful" phase of the algorithm. Previously, it would evict layers
   until the resident size fell below min_resident_size. In other words,
   we know protect one more layer of each tenant, so that the resident
   size stays just above min_resident_size, while previously we would
   evict enough to bring the resident size just under min_resident_size.

2. Previously, the "max layer size" that's used as the default
   min_resident_size was calculated from *all* layers in the tenant,
   including remote layers. Now it's only calculated across all
   locally-present layers. I don't know if that was a deliberate choice,
   but this is slightly simpler.
2023-03-28 01:28:27 +03:00
Christian Schwarz
0462563d31 doc: more explanation of the added config knobs 2023-03-27 19:52:48 +02:00
Christian Schwarz
ef39b3f067 rename: dedupe comment about allow(dead_code) 2023-03-27 19:27:41 +02:00
Christian Schwarz
e1ba2f96ab fix: post-merge, file_size() returns u64 2023-03-27 19:07:34 +02:00
Christian Schwarz
de3a7470de Merge remote-tracking branch 'origin/main' into problame/disk-usage-eviction 2023-03-27 19:04:55 +02:00
Christian Schwarz
504256a0cc rename: serde_percent::{Value => Percent} 2023-03-27 18:54:32 +02:00