Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

chore(ci): add model retry backoff configuration

Steven Enamakel41a88c0f687d
diff --git a/scripts/ci/rust-coverage.sh b/scripts/ci/rust-coverage.sh
index f7034a840e..0a994a2344 100755
--- a/scripts/ci/rust-coverage.sh
+++ b/scripts/ci/rust-coverage.sh
@@ -263,6 +263,13 @@ if [ "${OH_COV_RUNNER:-cargo}" = "nextest" ]; then
# process's counters online into a pool of 8 files per instrumented binary as
# it exits. cargo-llvm-cov reads the pattern from LLVM_PROFILE_FILE_NAME.
export LLVM_PROFILE_FILE_NAME="${LLVM_PROFILE_FILE_NAME:-openhuman-%8m.profraw}"
+ # The harness retries a failing model call on a 3/6/12/24 s schedule (about
+ # 45 s per turn). Suites that drive an unreachable or erroring upstream sat
+ # through it on a worker slot doing nothing: 9 of the 10 slowest tests in the
+ # warm EX63 run (60-90 s each) were that wait. The same schedule, with the
+ # waits scaled to 100 ms, keeps every retry and every assertion. The
+ # schedule itself is pinned by a unit test that ignores this variable.
+ export OPENHUMAN_MODEL_RETRY_BACKOFF_MS="${OPENHUMAN_MODEL_RETRY_BACKOFF_MS:-100}"
fi
log "running complete instrumented Rust suite (runner: ${OH_COV_RUNNER:-cargo})"