liran-funaro / fabric-x-committer / 29733006502

92%

main: 92%

push

github

[retry] Support unlimited retries via max-elapsed-time

Make retry.Profile.MaxElapsedTime a *time.Duration so an operator can
request unlimited retries, enabling dynamic scaling of pre-configured
service instances (the coordinator/sidecar keep reconnecting to
instances that are stopped/started at runtime).

Semantics:
  - nil (unset): default 15m (unchanged).
  - 0:           unlimited (never give up).
  - positive:    that budget.
A validate:"omitempty,gte=0" tag rejects negative values at config load.

Unlimited retries are provided by the backoff/Sustain layer
(WithMaxElapsedTime(0)). The gRPC retry policy has no notion of
unlimited attempts, so it stays intentionally bounded at
defaultGrpcMaxAttempts and relies on Sustain to re-establish the
connection; this applies to service reconnection only.

Also rewrite connection.CalcMaxAttempts as a closed-form formula
instead of a loop (behavior-preserving; validated by the unchanged
existing test vectors).

Document the dynamic-scaling retry-budget guidance and the 0=unlimited
option in the deployment guide and sample configs.

Relates to #679

Signed-off-by: Liran Funaro <liran.funaro@gmail.com>

26 of 26 new or added lines in 3 files covered. (100.0%)

10458 of 11410 relevant lines covered (91.66%)

43324.75 hits per line