Tobi Knaup created MESOS-1922:
---------------------------------
Summary: Slave blocks on the fetcher after terminating an executor
Key: MESOS-1922
URL: https://issues.apache.org/jira/browse/MESOS-1922
Project: Mesos
Issue Type: Bug
Reporter: Tobi Knaup
When the slave terminates an executor because the registration timeout hits, it
will hold on to the fetcher process if it is still running, and not send a
TASK_FAILED until the fetcher exists. Expected behavior would be to terminate
both the executor and the fetcher, and send then send the status update
immediately.
Here are some logs:
{code}
I1014 11:36:56.761726 209186816 slave.cpp:1139] Launching task
download.1370d754-53d1-11e4-9fc2-0a0027000000 for framework
20140927-211310-16777343-5050-44274-0001
I1014 11:36:56.766891 205430784 containerizer.cpp:394] Starting container
'be0f9918-986a-4692-ba9f-8c07871c5226' for executor
'download.1370d754-53d1-11e4-9fc2-0a0027000000' of framework
'20140927-211310-16777343-5050-44274-0001'
I1014 11:36:56.766922 209186816 slave.cpp:1252] Queuing task
'download.1370d754-53d1-11e4-9fc2-0a0027000000' for executor
download.1370d754-53d1-11e4-9fc2-0a0027000000 of framework
'20140927-211310-16777343-5050-44274-0001
I1014 11:36:56.768117 205430784 launcher.cpp:137] Forked child with pid '13624'
for container 'be0f9918-986a-4692-ba9f-8c07871c5226'
I1014 11:36:56.768647 205430784 containerizer.cpp:510] Fetching URIs for
container 'be0f9918-986a-4692-ba9f-8c07871c5226' using command
'/usr/local/libexec/mesos/mesos-fetcher'
I1014 11:37:43.375211 207577088 slave.cpp:3132] Current usage 92.50%. Max
allowed age: 0ns
I1014 11:37:56.768044 205967360 slave.cpp:3089] Terminating executor
download.1370d754-53d1-11e4-9fc2-0a0027000000 of framework
20140927-211310-16777343-5050-44274-0001 because it did not register within
1mins
I1014 11:37:56.768321 206503936 containerizer.cpp:882] Destroying container
'be0f9918-986a-4692-ba9f-8c07871c5226'
I1014 11:37:56.817491 207577088 containerizer.cpp:997] Executor for container
'be0f9918-986a-4692-ba9f-8c07871c5226' has exited
{code}
At this point there is still a running fetcher. After killing it manually I see:
{code}
W1014 11:49:06.310417 207040512 containerizer.cpp:872] Ignoring destroy of
unknown container: be0f9918-986a-4692-ba9f-8c07871c5226
E1014 11:49:06.310560 208650240 slave.cpp:2564] Container
'be0f9918-986a-4692-ba9f-8c07871c5226' for executor
'download.1370d754-53d1-11e4-9fc2-0a0027000000' of framework
'20140927-211310-16777343-5050-44274-0001' failed to start: Failed to fetch
URIs for container 'be0f9918-986a-4692-ba9f-8c07871c5226': exit status 15
E1014 11:49:06.310597 208650240 slave.cpp:2659] Termination of executor
'download.1370d754-53d1-11e4-9fc2-0a0027000000' of framework
'20140927-211310-16777343-5050-44274-0001' failed: Unknown container:
be0f9918-986a-4692-ba9f-8c07871c5226
E1014 11:49:06.310699 205430784 slave.cpp:2945] Failed to unmonitor container
for executor download.1370d754-53d1-11e4-9fc2-0a0027000000 of framework
20140927-211310-16777343-5050-44274-0001: Not monitored
I1014 11:49:06.315104 208650240 slave.cpp:2115] Handling status update
TASK_FAILED (UUID: c216eb10-cfdc-4a9e-a687-e260701daed4) for task
download.1370d754-53d1-11e4-9fc2-0a0027000000 of framework
20140927-211310-16777343-5050-44274-0001 from @0.0.0.0:0
W1014 11:49:06.315213 208113664 containerizer.cpp:788] Ignoring update for
unknown container: be0f9918-986a-4692-ba9f-8c07871c5226
I1014 11:49:06.315398 205967360 status_update_manager.cpp:320] Received status
update TASK_FAILED (UUID: c216eb10-cfdc-4a9e-a687-e260701daed4) for task
download.1370d754-53d1-11e4-9fc2-0a0027000000 of framework
20140927-211310-16777343-5050-44274-0001
I1014 11:49:06.315489 205967360 status_update_manager.cpp:373] Forwarding
status update TASK_FAILED (UUID: c216eb10-cfdc-4a9e-a687-e260701daed4) for task
download.1370d754-53d1-11e4-9fc2-0a0027000000 of framework
20140927-211310-16777343-5050-44274-0001 to [email protected]:5050
I1014 11:49:06.328732 209186816 status_update_manager.cpp:398] Received status
update acknowledgement (UUID: c216eb10-cfdc-4a9e-a687-e260701daed4) for task
download.1370d754-53d1-11e4-9fc2-0a0027000000 of framework
20140927-211310-16777343-5050-44274-0001
I1014 11:49:06.328951 207040512 slave.cpp:2811] Cleaning up executor
'download.1370d754-53d1-11e4-9fc2-0a0027000000' of framework
20140927-211310-16777343-5050-44274-0001
{code}
To reproduce, just launch a task with a URI that takes longer than executor
registration timeout to download.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
