-
Notifications
You must be signed in to change notification settings - Fork 3
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
Add batched exports to the Prometheus Remote Write Exporter #34
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM just some small suggestions about permanent errors
var wg sync.WaitGroup | ||
var mtx sync.Mutex | ||
|
||
for _, request := range requests { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One thing I'm worried about this is it can create a large number of concurrent outgoing request. We should cap the max number of concurrent requests.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Added a cap on the number of concurrent requests (with some logs) and addressed the comments you suggested above. Here's an example of the logs produced with this commit where the cap was 3 concurrent requests.
total metrics to process: 194128
tsArray length: 44566
sizeOfCurrentBatch: 2999930
tsArray length: 44560
sizeOfCurrentBatch: 2999986
tsArray length: 44555
sizeOfCurrentBatch: 2999962
tsArray length: 44553
sizeOfCurrentBatch: 2999978
tsArray length: 15894
sizeOfCurrentBatch: 1069220
next batch of requests
making request 0
making request 1
making request 2
next batch of requests
making request 4
making request 3
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Left one comment, otherwise LGTM.
var mu sync.Mutex | ||
numOfRequests := len(requests) | ||
|
||
for i := 0; i < numOfRequests; i += maxConcurrentRequests { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This will wait for the slowest request to be completed before starting to consume the next. We can use a buffered channel instead.
An example of this pattern is https://github.com/google/gops/blob/master/goprocess/gp.go#L55
Not sure if it's readable enough, let me know if you need help.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Ohh, I see! I'll try that out
LGTM |
This PR has been made upstream. |
* Sync build and CI Go versions at latest 1.16 * Run go mod tidy * Set go binary to use in the compilation phase in tests Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> Co-authored-by: Juraci Paixão Kröhling <juraci@kroehling.de>
* Initial commit * Add CODEOWNERS file (#2) * Add CODEOWNERS file * Update CODEOWNERS * Moved from github.com/observatorium/opentelemetry-collector-builder (#3) Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> * fixed panics (#6) Signed-off-by: Joe Elliott <number101010@gmail.com> * Replace master with main in CI and mergify files (#8) Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Bump to OpenTelemetry Collector 0.20.0 (#10) Closes #9 Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Explicitly enable Go modules in quickstart instructions (#13) * Update to collector v0.21.0 (#17) Fixes #16 Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Update to collector v0.22.0 (#19) * Download go modules before building (#20) Fixes #14 * Add version command (#25) Signed-off-by: Ashmita Bohara <ashmita.bohara152@gmail.com> * Pass errors from cobra Execute back to main for correct exit code (#28) * pass errors from cobra execute back to main * print the error * Update to collector v0.23.0 (#27) * Generate a warning if the builder and collector base version mismatch (#30) * Generate a warning if the builder and collector base version mismatch * Show current default version in the warning message * Update to OpenTelemetry Collector 0.24.0 * Don't use %w formatting with log.Fatal (#35) * Update to OpenTelemetry Collector 0.25.0 (#36) Signed-off-by: Serge Catudal <serge.catudal@gmail.com> * Update to 0.26.0 and update BuildInfo (#39) * Sync build and CI Go versions at latest 1.16 (#34) * Sync build and CI Go versions at latest 1.16 * Run go mod tidy * Set go binary to use in the compilation phase in tests Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> Co-authored-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Add option to generate go code only (no compile) (#40) * Issue#24 Add option to generate go code only (no compile) * Update cmd/root.go logging Suggested by @jpkkrohling Co-authored-by: Juraci Paixão Kröhling <juraci.github@kroehling.de> * remove verbose help .. created by corba * suggestion by jpkrohling to keep generateandcompile * lint error: remove unused var * reword cmd option and add back help message for default Co-authored-by: Juraci Paixão Kröhling <juraci.github@kroehling.de> * Don't reuse exec.Cmd (#42) * Update to OpenTelemetry Collector 0.27.0 (#43) * Add CI Badge (#47) * Update to Collector v0.28.0 (#49) * Update to Collector v0.28.0 Closes #48 Addresses the breaking API change in open-telemetry#3163, besides the usual version number changes. Signed-off-by: Fangyi Zhou <me@fangyi.io> * Use `go mod tidy` instead of `go mod download` It appears that this magically resolves the go.mod file issue. https://stackoverflow.com/questions/67203641/missing-go-sum-entry-for-module-providing-package-package-name Signed-off-by: Fangyi Zhou <me@fangyi.io> * Account for go mod download in go1.17 not updating go.sum (#50) * Update to collector v0.29.0 (#54) * Update replaces.builder.yaml * Update nocore.builder.yaml * Update config.go * Update README.md * Update main.go * Update to collector v0.30.0 (#57) * cmd: fix module flag default value to github.com/open-telemetry (#58) Signed-off-by: Koichi Shiraishi <zchee.io@gmail.com> * Update to collector v0.31.0 (#60) * Update to v0.33.0 (#62) Signed-off-by: Anthony J Mirabella <a9@aneurysm9.com> * Add excludes support to generated go.mod (#63) Signed-off-by: Anthony J Mirabella <a9@aneurysm9.com> Co-authored-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Small cleanup for the builder files (#64) Signed-off-by: Bogdan Drutu <bogdandrutu@gmail.com> * Support building with Go 1.17 (#66) * Support building with Go 1.17 Fixes #65 Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Update workflows to use Go 1.17 Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Add gosec exceptions for exec.Command Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Update to OpenTelemetry core 0.34.0 (#68) Fixes #67 Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> * Upgrade to OpenTelemetry Collector 0.35.0 (#70) Signed-off-by: Fangyi Zhou <me@fangyi.io> * Upgrade to OpenTelemetry Collector 0.36.0 (#76) * Generate custom service code for Windows (#75) * update main to include windows service code * use main version from tag 0.35.0 * update main function * align with upstream v0.36.0 tag * dummy change to trigger build * Revert "dummy change to trigger build" This reverts commit 629d499461da2d2c240bf1e495b5fe0558e3547f. * Remove Core from Module type (#77) Fixes #15 Signed-off-by: yugo-horie <u5.horie@gmail.com> * release 0.37.0 (#78) * release 0.37.0 * update use of NewCommand * Move builder to subdirectory Signed-off-by: Juraci Paixão Kröhling <juraci@kroehling.de> Co-authored-by: Bogdan Drutu <lazy@splunk.com> Co-authored-by: Bogdan Drutu <bogdandrutu@gmail.com> Co-authored-by: Joe Elliott <joe.elliott@grafana.com> Co-authored-by: Eric Yang <jiwen624@gmail.com> Co-authored-by: Brian Gibbins <eroteme@supernought.co.uk> Co-authored-by: Ashmita <ashmita.bohara152@gmail.com> Co-authored-by: Fangyi Zhou <me@fangyi.io> Co-authored-by: Shaun Creary <65406540+crearys@users.noreply.github.com> Co-authored-by: Patryk Małek <69143962+pmalek-sumo@users.noreply.github.com> Co-authored-by: Serge Catudal <serge.catudal@gmail.com> Co-authored-by: Aaron Stone <aaron@serendipity.cx> Co-authored-by: Patryk Małek <pmalek@sumologic.com> Co-authored-by: Aaron Stone <aaron.stone@udacity.com> Co-authored-by: Kelvin Lo <kello@live.ca> Co-authored-by: Himanshu <addyjeridiq@gmail.com> Co-authored-by: Y.Horie <u5.horie@gmail.com> Co-authored-by: Koichi Shiraishi <zchee.io@gmail.com> Co-authored-by: Anthony Mirabella <a9@aneurysm9.com> Co-authored-by: Cal Loomis <68860480+loomis-relativity@users.noreply.github.com> Co-authored-by: alrex <aboten@lightstep.com>
Description:
Currently, there is no limit on the size of the remote_write export request to a Cortex instance. This could create problems if one Prometheus scrape produces an extreme amount of timeseries metrics.
This PR creates a batch of concurrent requests where the max size of a request is 3MB and the max number of timeseries per request is 40,000.
Testing: