Repository navigation
Honor Retry-After on throttled Azure blob responses - #2931
Merged
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
amankrx
force-pushed
the
fix/azure-throttle-retry
branch
from
October 13, 2026 14:15
5e04775 to
bbe74eb
Compare
corcillo
reviewed
Oct 13, 2026
corcillo
reviewed
Oct 13, 2026
corcillo
previously approved these changes
Oct 13, 2026
…t outside the retry attempts
corcillo
approved these changes
Oct 13, 2026
This was referenced Oct 13, 2026
MarcusSorealheis
added a commit
to corcillo/nativelink
that referenced
this pull request
Oct 14, 2026
Resolve conflicts with Retry-After support (TraceMachina#2931): the retrier keeps both its client label and outcome hook and the throttle budget, and Azure sets both. A throttle waited out within the budget reports `retried`, and a success after one reports `recovered`, though it spends no attempt. Regenerate the metrics reference.
b7r6
added a commit
to b7r6/nativelink
that referenced
this pull request
Oct 14, 2026
…pl added upstream (TraceMachina#2931) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This branch was successfully deployed
This file contains hidden or 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 and why
A storage account over its egress limit answers 503 ServerBusy for minutes, and the Azure store treated that like any transient error: the retrier's budget (about 38 s with 6 retries from 0.3 s) ran out, the action failed and was re-dispatched to fetch the same blobs again. One customer's workers failed 786,837 input fetches that way in 18 hours, all on the seventh attempt.
The Azure store classifies 503 ServerBusy and 429 as throttles and reads
retry-after-ms,x-ms-retry-after-msorretry-after. The retrier gainsRetryResult::RetryAfter(err, wait)and a throttle budget: withinthrottle_retry_budget_s(default 120) of one operation each throttle waits what the server asked, betweenthrottle_retry_delay_s(2) andthrottle_retry_max_delay_s(60, a longer header is capped and logged), and spends none ofretry.max_retries. Past the budget, throttles spend attempts and backoff like any other error, so a throttle that never ends still fails. Counterstore.object.throttles{provider,op}.How was this verified?
Retrier tests for the hint and the backoff, throttles inside and past the budget, and a zero wait; Azure store tests under paused time for the header, the floor, the cap and three throttles on zero allowed retries. nativelink-util, config and store suites and clippy pass.
Risk
Low. Non-throttled errors keep their path; a throttled fetch holds its slot for up to the budget, bounded.
AI assistance
An agent drafted the change and I reviewed every line.