hydra: raise maxUnsupportedTimeInS to 24h
Builders register over gRPC, so a reconnect or network blip briefly drops their system from the queue runner. With the 120s default, abort_unsupported then fails buildable steps as "Unsupported system type" (NixOS/hydra#1805).
Jörg Thalheim committed
Jun 18, 2026 at 10:12 UTC
707f96d4e82809aa0422add75c70a04fe985e500
2 files changed
+6
build/hydra-queue-runner.nix
+3
@@ -89,6 +89,9 @@ in
89
maxOutputSize = 4294967295; # 4 GiB - 1 B, matches prod hydra.conf max_output_size
90
# TODO: Expose dispatchTriggerTimerInS, defaults to 120s
91
queueTriggerTimerInS = 60;
92
+ # bump from the 120s default: builder reconnects briefly drop their
93
+ # system and we'd abort buildable steps as unsupported (hydra#1805)
94
+ maxUnsupportedTimeInS = 86400;
95
concurrentUploadLimit = 48;
96
maxConcurrentDownloads = 48;
97
remoteStoreAddr = [
non-critical-infra/hosts/staging-hydra/hydra.nix
+3
@@ -120,6 +120,9 @@ in
120
settings = {
121
queueTriggerTimerInS = 300;
122
concurrentUploadLimit = 2;
123
+ # bump from the 120s default: builder reconnects briefly drop their
124
+ # system and we'd abort buildable steps as unsupported (hydra#1805)
125
+ maxUnsupportedTimeInS = 86400;
126
remoteStoreAddr = [
127
"s3://nix-cache-staging?secret-key=${config.sops.secrets.signing-key.path}&ls-compression=br&log-compression=br"
128
];