Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

WIP: Not for merging and please ignore. #8996

Closed
wants to merge 1 commit into from

Conversation

gotjosh
Copy link
Contributor

@gotjosh gotjosh commented Aug 14, 2024

What this PR does

N/A only here for illustrative purposes

Which issue(s) this PR fixes or relates to

N/A

Checklist

  • Tests updated.
  • Documentation added.
  • CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX].
  • about-versioning.md updated with experimental features.

@gotjosh gotjosh force-pushed the dimitar/ingester/consume-latency-push-sharding branch from 05eae9e to 0d4a616 Compare August 29, 2024 09:19
@dimitarvdimitrov dimitarvdimitrov force-pushed the dimitar/ingester/consume-latency-push-sharding branch 2 times, most recently from c0d740c to fb30f39 Compare September 20, 2024 13:12
@gotjosh gotjosh force-pushed the dimitar/ingester/consume-latency-push-sharding branch from fc89859 to 861563d Compare September 26, 2024 17:14
@dimitarvdimitrov dimitarvdimitrov force-pushed the dimitar/ingester/consume-latency-push-sharding branch from 0cb28d6 to 23d4e28 Compare September 27, 2024 12:12
Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

Use labels hasher

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

Use consistent title name

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

Use consistent title name

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

kafka replay speed: adjust batchingQueueCapacity (#9344)

* kafka replay speed: adjust batchingQueueCapacity

I made 2000 up when we were flushing individual series to the channel.
Then 2000 might have made sense, but when flushing whole WriteRequests
a capacity of 1 should be sufficient.

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Increase errCh capacity

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Explain why +1

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Set capacity to 5

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Update pkg/storage/ingest/pusher.go

Co-authored-by: gotjosh <josue.abreu@gmail.com>

* Improve test

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Update pkg/storage/ingest/pusher.go

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
Co-authored-by: gotjosh <josue.abreu@gmail.com>

kafka replay speed: rename CLI flags (#9345)

* kafka replay speed: rename CLI flags

Make them a bit more consistent on what they mean and add better descriptions.

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Clarify metrics

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Rename flags

Co-authored-by: gotjosh <josue.abreu@gmail.com>

* Update docs

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
Co-authored-by: gotjosh <josue.abreu@gmail.com>

kafka replay speed: add support for metadata & source (#9287)

* kafka replay speed: add support for metadata & source

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Remove completed TODO

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Use a single map

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Make tests compile again

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

kafka replay speed: improve fetching tracing (#9361)

* Better span logging

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Better span logging

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Try to have more buffering in ordered batches

maybe waiting to send to ordered batches comes with too much overhead

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Correct local docker-compose config with new flags

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Maybe have more stable events

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Revert "Try to have more buffering in ordered batches"

This reverts commit 886b159.

* Maybe have more stable events

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Maybe have more stable events

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Propagate loggers in spans

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

continuous-test: Make the User-Agent header for the Mimir client conf… (#9338)

* continuous-test: Make the User-Agent header for the Mimir client configurable

* Update CHANGELOG.md

* Run make reference-help

TestIngester_PushToStorage_CircuitBreaker: increase initial delay (#9351)

* TestIngester_PushToStorage_CircuitBreaker: increase initial delay

Fixes XXX
I believe there's a race between sending the first request and then collecting the metrics. It's possible that we collect the metrics longer than 200ms after the first request, at which point the CB has opened. I could reproduce XXX by reducing the initialDelay to 10ms.

This PR increases it to 1 hour so that we're more sure that the delay hasn't expired when we're collecting the metrics.

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Adjust all tests

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

Update to latest commit of dskit main (#9356)

Specifically pulls in grafana/dskit#585

Signed-off-by: Nick Pillitteri <nick.pillitteri@grafana.com>

Update mimir-prometheus (#9358)

* Update mimir-prometheus

* Run make generate-otlp

query-tee: add equivalent errors for string expression for range queries (#9366)

* query-tee: add equivalent errors for string expression for range queries

* Add changelog entry

MQE: fix `rate()` over native histograms where first point in range is a counter reset (#9371)

* MQE: fix `rate()` over native histograms where first point is a counter reset

* Add changelog entry

Update module github.com/Azure/azure-sdk-for-go/sdk/storage/azblob to v1.4.1 (#9369)

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Use centralized 'Add to docs project' workflow with GitHub App auth (#9330)

* Use centralized 'Add to docs project' workflow with GitHub App auth

Until this is merged, it is likely that any issues labeled `type/docs` won't be added to the [organization project](https://github.com/orgs/grafana/projects/69).

The underlying action is centralized so that any future changes are made in one place (`grafana/writers-toolkit`). The action is versioned to protect workflows from breaking changes.

The action uses Vault secrets instead of the discouraged organization secrets.

The workflow uses a consistent name so that future changes can be made programmatically.

Relates to https://github.com/orgs/grafana/projects/279/views/9?pane=issue&itemId=44280262

Signed-off-by: Jack Baldry <jack.baldry@grafana.com>

* Remove unneeded checkout step

* Remove unneeded checkout step

---------

Signed-off-by: Jack Baldry <jack.baldry@grafana.com>

Update grafana/agent Docker tag to v0.43.1 (#9365)

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Update module github.com/hashicorp/vault/api/auth/userpass to v0.8.0 (#9375)

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Update module github.com/hashicorp/vault/api/auth/approle to v0.8.0 (#9374)

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Update module go.opentelemetry.io/collector/pdata to v1.15.0 (#9380)

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Update module github.com/hashicorp/vault/api/auth/kubernetes to v0.8.0 (#9377)

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

Update module github.com/twmb/franz-go/plugin/kotel to v1.5.0 (#9379)

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

kafka replay speed: ingestion metrics (#9346)

* kafka replay speed: ingestion metrics

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Separate batch processing time by batch contents

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Also set time on metadata

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Add tenant to metrics

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Add metrics for errors

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Rename batching queue metrics

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Pairing to address code review

Co-Authored-By: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Move the metrics into their own file

Co-Authored-By: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* go mod tidy

Signed-off-by: gotjosh <josue.abreu@gmail.com>

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
Signed-off-by: gotjosh <josue.abreu@gmail.com>
Co-authored-by: gotjosh <josue.abreu@gmail.com>

kafka replay speed: move error handling closer to actual ingestion (#9349)

* kafka replay speed: move error handling closer to actual ingestion

Previously, we'd let error bubble-up and only take decisions on whether to abort the request or not at the very top (`pusherConsumer`). This meant that we'd potentially buffer more requests before we detect an error.

This change extracts error handling logic into a `Pusher` implementation: `clientErrorFilteringPusher`. This implementation logs client errors and then swallows them. We inject that implementation in front of the ingester. This means that the parallel storage implementation can abort ASAP instead of collecting and bubbling up the errors.

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
Signed-off-by: gotjosh <josue.abreu@gmail.com>
Co-authored-by: gotjosh <josue.abreu@gmail.com>

kafka replay speed: concurrency fetching improvements (#9389)

* fetched records include timestamps

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* try with defaultMinBytesWaitTime=3s

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* add fetch_min_bytes_max_wait

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Don't block on sending to the channel

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Remove wait for when we're fetching from the end

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Fix bug with blocking on fetch

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Slightly easier to follow lifecycle of previousResult

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Correct merging of results

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Avoid double-logging events

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Revert "add fetch_min_bytes_max_wait"

This reverts commit 6197d4b.

* Increase MinBytesWaitTime to 5s

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Add comment about warpstream and MinBytes

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Address review comments

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* Add tests for concurrentFetchers

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Fix bugs in tracking lastReturnedRecord

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Renamed method

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* use the older context

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* Name variable correct variable name

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Reduce MaxWaitTime in PartitionReader tests

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Change test createConcurrentFetchers signature

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Sort imports

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
Signed-off-by: gotjosh <josue.abreu@gmail.com>
Co-authored-by: gotjosh <josue.abreu@gmail.com>

Make concurrentFetchers change its concurrency dynamically (#9437)

* Make concurrentFetchers change its concurrency dynamically

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* address review comments

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* `make doc`

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* inline the stop method

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* Fix panic when creating concurrent fetchers fails

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Disabled by default

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* we don't need to handle the context in start

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* don't store concurrency or records per fetch

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* add validation to the flags

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* Ensure we don't leak any goroutines.

Signed-off-by: gotjosh <josue.abreu@gmail.com>

* remove concurrent and recordsperfetch from the main struct

Signed-off-by: gotjosh <josue.abreu@gmail.com>

---------

Signed-off-by: gotjosh <josue.abreu@gmail.com>
Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
Co-authored-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

kafka replay speed: fix concurrent fetching concurrency transition (#9447)

* kafka replay speed: fix concurrent fetching concurrency transition

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Update pkg/storage/ingest/reader.go

* Make sure we evaluate r.lastReturnedRecord WHEN we return

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Redistribute wg.Add

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Add tests

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Remove defer causing data race

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Move defer to a different place

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* WIP

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Give more time to catch up with target_lag

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

* Clarify comment

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>

---------

Signed-off-by: Dimitar Dimitrov <dimitar.dimitrov@grafana.com>
@dimitarvdimitrov dimitarvdimitrov force-pushed the dimitar/ingester/consume-latency-push-sharding branch from 23d4e28 to 95f063e Compare September 27, 2024 12:15
@gotjosh gotjosh closed this Sep 30, 2024
@gotjosh gotjosh deleted the dimitar/ingester/consume-latency-push-sharding branch September 30, 2024 16:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants