fix(blooms): Match series to newest block only #15481
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
based on meta's TSDB timestamp Signed-off-by: Christian Haudum <christian.haudum@gmail.com>
Signed-off-by: Christian Haudum <christian.haudum@gmail.com>
chaudum
commented
Dec 18, 2024
Comment on lines
+112
to
+117
| for i := range res { | ||
| t.Logf("%s", res[i].block) | ||
| for j := range res[i].series { | ||
| t.Logf(" %016x", res[i].series[j].Fingerprint) | ||
| } | ||
| } |
There was a problem hiding this comment.
mainly for debugging
salvacorts
reviewed
Dec 19, 2024
salvacorts
reviewed
Dec 19, 2024
Signed-off-by: Christian Haudum <christian.haudum@gmail.com>
The the slice of sources must be the same length as the slice of blocks. A source at index i is the TSDB identifier from which the block at index i was built. Signed-off-by: Christian Haudum <christian.haudum@gmail.com>
Signed-off-by: Christian Haudum <christian.haudum@gmail.com>
salvacorts
approved these changes
Dec 19, 2024
This was referenced Dec 23, 2024
mveitas
pushed a commit
to mveitas/loki
that referenced
this pull request
Jan 6, 2025
**What this PR does / why we need it**:
While running bloom filters in production we noticed some Loki clusters that showed a very high percentage of missing chunks when querying blooms, thus resulting in lower filter rate.
The reason is that old, superseded blocks are still considered up-to-date, because they cover a keyspace that is not covered by newer blocks with smaller keyspaces (due to larger individual series).
```
| series fingerprint keyspace
------------+----------------------------------------------------------------
| o o o o o o o o o o o o o
------------+----------------------------------------------------------------
iteration 1 | 111111111111111111111111111111111111111111111111111111111
iteration 2 | 22222222222222 3333333333333333 444444
iteration 3 | 5555555 6666666 77777777 888888888 9999999999
...
up-to-date | 555555522266666661111177777777333388888888811119999999999
------------+----------------------------------------------------------------
| x
```
The chart shows the different blocks marked with the numbers 1 to 9 for a subset of the full series fingerprint keyspace. The blocks are generated in multiple successive bloom building iterations. The first block covers a larger keyspace (more series), because the individual blooms in the blocks are smaller in the beginning of the day. Later, the blooms get larger and therefore the block fingerprint ranges gets smaller. However, since we are dealing with fingerprint ranges, not individual fingerprints, the newer blocks cause "gaps" in the range of the previously larger keyspace. In the case above, every block except block 4, are considered up-to-date, since each of them covers a keyspace that is otherwise not covered.
When resolving blocks for a series at query time, we consider looking at all up-to-date blocks, which are referenced by the meta files. The series `x` in the chart shows, that it is within the range of 3 up-to-date blocks: 1, 2, 5. However, only the newest block (5) may contain the requested series.
This PR changes the block resolver on the index-gateway to only match the newest block to a series, based on the timestamp of the TSDB from with the blocks were generated.
---
Signed-off-by: Christian Haudum <christian.haudum@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What this PR does / why we need it:
While running bloom filters in production we noticed some Loki clusters that showed a very high percentage of missing chunks when querying blooms, thus resulting in lower filter rate.
Screenshot from Bloom Gateway dashboard
The reason is that old, superseded blocks are still considered up-to-date, because they cover a keyspace that is not covered by newer blocks with smaller keyspaces (due to larger individual series).
The chart shows the different blocks marked with the numbers 1 to 9 for a subset of the full series fingerprint keyspace. The blocks are generated in multiple successive bloom building iterations. The first block covers a larger keyspace (more series), because the individual blooms in the blocks are smaller in the beginning of the day. Later, the blooms get larger and therefore the block fingerprint ranges gets smaller. However, since we are dealing with fingerprint ranges, not individual fingerprints, the newer blocks cause "gaps" in the range of the previously larger keyspace. In the case above, every block except block 4, are considered up-to-date, since each of them covers a keyspace that is otherwise not covered.
When resolving blocks for a series at query time, we consider looking at all up-to-date blocks, which are referenced by the meta files. The series
xin the chart shows, that it is within the range of 3 up-to-date blocks: 1, 2, 5. However, only the newest block (5) may contain the requested series.This PR changes the block resolver on the index-gateway to only match the newest block to a series, based on the timestamp of the TSDB from with the blocks were generated.