cloud_storage: use remote index in cloud timequery #13011
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
A lookup by timestamp is added to the remote index. It has the same semantics as the other lookup methods. If the index does not include the time index (i.e. it was created with serde version 1), the search comes up empty.
VladLazar
force-pushed
the
use-index-in-cloud-timequery
branch
from
ea3677a to
78454bf
Compare
VladLazar
force-pushed
the
use-index-in-cloud-timequery
branch
from
78454bf to
0e88306
Compare
|
Changes in force-push:
|
dotnwat
reviewed
Aug 25, 2023
| @@ -184,6 +184,58 @@ offset_index::find_kaf_offset(kafka::offset upper_bound) { | |||
| return res; | |||
| } | |||
|
|
|||
| std::optional<offset_index::find_result> | |||
| offset_index::find_timestamp(model::timestamp upper_bound) { | |||
| if (_initial_time == model::timestamp::missing()) { | |||
There was a problem hiding this comment.
it's a bummer that -1 is used to indicate missing timestamp. That -1 carries so much additional baggage in Kafka for example by indicating something specific in time queries.
There was a problem hiding this comment.
Yeah, it's not ideal. Not much we can do about it now though. The -1 is already encoded in a released Serde version.
| @@ -315,18 +315,18 @@ remote_segment::offset_data_stream( | |||
| offset_index::find_result pos; | |||
| std::optional<uint16_t> prefetch_override = std::nullopt; | |||
| if (first_timestamp) { | |||
There was a problem hiding this comment.
is it odd that first_timestamp wasn't used in this conditional before?
There was a problem hiding this comment.
first_timestamp indicates that this is a time-query. By this point, we have already resolved to the correct segment for the time query. Since we didn't previously used the index, it makes senes that first_timestamp wasn't used (we'd just start from the beginning of the segment).
abhijat
reviewed
Aug 28, 2023
abhijat
reviewed
Aug 28, 2023
added 3 commits
August 29, 2023 11:26
This commit updates the timequery read path to skip up to the first index entry with a timestamp smaller than the searched one. The result is that timequeries will hydrate/materialize a maximum of two chunks (two because the chunk boundaries don't always line up with the entries in the index). If the index is not present, or was originally created in v1, the search starts from the first chunk as it did previously.
A shard-level metric is added to track the total number of chunks that were hydrated (i.e. downloaded).
This commit makes a couple of changes to the timequery tests: 1. Run timequery on more offests (10 for each of the 12 segments) 2. Check that a maximum of two chunks are downloaded by any given timequery 3. Use the admin api to get the precise boundary between the cloud and local log. Previously, it was estimated based on record size.
VladLazar
force-pushed
the
use-index-in-cloud-timequery
branch
2 times, most recently
from
604d114 to
c5df897
Compare
VladLazar
force-pushed
the
use-index-in-cloud-timequery
branch
from
c5df897 to
6314700
Compare
This commit refactors the handling of the index search result. If a new result type is introduced, it's author will be reminded to handle it by the assertions.
VladLazar
force-pushed
the
use-index-in-cloud-timequery
branch
from
6314700 to
de5aa53
Compare
abhijat
approved these changes
Aug 29, 2023
|
/backport v23.2.x |
|
Failed to run cherry-pick command. I executed the commands below: |
|
/backport v23.2.x |
|
Failed to run cherry-pick command. I executed the commands below: |
7 tasks
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.
This PR updates the timequery read path to skip up to the first
index entry with a timestamp smaller than the searched one. The result
is that timequeries will hydrate/materialize a maximum of two chunks
(two because the chunk boundaries don't always line up with the entries
in the index).
If the index is not present, or was originally created in v1, the search
starts from the first chunk as it did previously.
Fixes #11801
Backports Required
Release Notes
Improvements
and reduce the number of hydrated bytes required to serve the query. On average, a time query will have to download 4 times less data (if using the default segment and chunk size)