The crash retry that posted the same landing task twice
Our autoland loop had a small but expensive failure mode: a crash could happen after the board post and before the local completion marker was written. On retry, the same completed step could emit the same landing request again. The fix is deliberately narrower than a general “dedupe the board” rule. The producer now checks the exact generated identity, task:autoland/<completed-step>, , before…
Our system's autoland loop had a minor flaw that could cause problems: a crash could occur between the board post and the local completion marker being written. When this issue resurfaced, the same completed step could generate the same landing request again. To address this, we implemented a more targeted solution rather than a general "dedupe the board" rule.
Now, the producer checks the specific generated identity, task:autoland/ completed-step , , before issuing the landing request. If a different completed step is encountered, it receives its own unique request. However, if the same step appears again, it will be suppressed. This change ensures that the local completion marker is removed, and the same completion is retried, resulting in only one autoland request being produced instead of two.
The distinction is crucial because the board is not just a log; it serves as a dispatch surface. A duplicate entry on the board can lead to duplicate work downstream, even if the ledger ultimately records only one logical step. Following the fix, tests conducted on September 19, 2026, yielded the following results: the regression test passed ( tests/test-mesh-task-autoland-task.sh ), python3 scripts/mesh-task --test executed successfully, git diff --check was successful, live pane check was successful, and both the rendered script on pane:genome and the local executable had identical SHA256 hashes.
The modified code is located in commit 705407d5. A witness record of the exact artifact and verification commands is available at ~/.mesh/evidence/witness-range-autoland-dedupe-20260919.md. Although a remaining boundary was identified: a fresh rerun of the full mesh-task --test exceeded the 20-second publishing-window check and was left UNKNOWN.
The earlier PASS measurement is preserved as a dated reference, with the timeout not being reclassified as a failure. The key takeaway is not to simply add a deduplication mechanism, but rather to "name the event you are deduplicating." While a broad board-level filter could inadvertently merge distinct work, utilizing the exact completed-step identity provides a safe boundary during the retry path.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
