Skip to content

ate-setup: settings for the router's route timeout, parked budget and ext_proc breaker - #2560

Open
Davanum Srinivas (dims) wants to merge 2 commits into
agent-substrate:mainfrom
dims:router-settings
Open

Davanum Srinivas (dims) wants to merge 2 commits into
agent-substrate:mainfrom
dims:router-settings

Conversation

@dims

Copy link
Copy Markdown
Collaborator

The router flags --route-timeout, --parked-request-budget and --extproc-max-requests could only be set by a patch on the atenet-router deployment. At scale, the 5 s default parked budget and the derived ext_proc circuit breaker were the first limits hit.

This adds three install settings, ATE_ATENET_ROUTE_TIMEOUT, ATE_ATENET_PARKED_REQUEST_BUDGET and ATE_ATENET_EXTPROC_MAX_REQUESTS, each with a flag and a config-file form. The installer validates them (positive values, envoy dataplane only), sets the flags on the router container in both deploy paths, grows the router's terminationGracePeriodSeconds with the budget, and documents them.

Verified with gofmt -l on the touched Go files, go vet ./cmd/ate-setup/..., go test ./cmd/ate-setup/... (including the doc-drift tests) and hack/verify/boilerplate.sh.

@dims
Davanum Srinivas (dims) force-pushed the router-settings branch 2 times, most recently from bf2a288 to 66dd6eb Compare October 14, 2026 12:08
@bowei Bowei Du (bowei) added area/dev-infra area/network kind/feature An enhancement / feature request or implementation labels Oct 14, 2026
@dims
Davanum Srinivas (dims) force-pushed the router-settings branch 6 times, most recently from 9d95687 to 4c2b4d2 Compare October 22, 2026 12:37
@dims
Davanum Srinivas (dims) force-pushed the router-settings branch 2 times, most recently from ee6dc28 to 968f0ca Compare October 23, 2026 15:04
--atenet-route-timeout and --atenet-parked-request-budget set the two
router flags at install time; unset keeps the router defaults. A larger
budget grows terminationGracePeriodSeconds by the same amount so parked
requests finish during a rollout. Envoy dataplane only.

applyWithAPIServerEnvHash now takes the decoded objects, so the system
bundle is decoded once, the router flags are set, and the api-server
env hash is stamped before the one apply.
ATE_ATENET_EXTPROC_MAX_REQUESTS (atenet.router.extprocMaxRequests,
--atenet-extproc-max-requests) passes --extproc-max-requests to the
router the way the route timeout and parked budget settings pass
theirs. On a dense host 1536 synchronized execs overflowed the derived
breaker of 2048 and Envoy answered 503 "reset reason: overflow"; the
only way to raise it was a patch on the deployment.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area/dev-infra area/network kind/feature An enhancement / feature request or implementation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants