ExternalVendorCode/onefuzz
1
0
Fork 0
mirror of https://github.com/microsoft/onefuzz.git synced 2025-06-17 04:18:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

mark tasks as failed when the node is being reimaged due to heartbeat issues (#1015)

Browse Source
This commit is contained in:
bmc-msft
2021-06-23 16:39:47 -04:00
committed by GitHub
parent 97cba8eda1
commit 50652c2e48
1 changed files with 10 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/api-service/__app__/onefuzzlib/workers/scalesets.py
View File

@ -391,6 +391,16 @@ class Scaleset(BASE_SCALESET, ORMMixin):
",".join(str(x.machine_id) for x in dead_nodes), ",".join(str(x.machine_id) for x in dead_nodes),
) )
for node in dead_nodes: for node in dead_nodes:
error = Error(
code=ErrorCode.TASK_FAILED,
errors=[
"node reimaged due to expired heartbeat",
f"scaleset_id:{node.scaleset_id} machine_id:{node.machine_id}",
f"last heartbeat:{node.heartbeat}",
],
)
node.mark_tasks_stopped_early(error=error)
node.to_reimage(done=True)
if node not in to_reimage: if node not in to_reimage:
to_reimage.append(node) to_reimage.append(node)