-
Notifications
You must be signed in to change notification settings - Fork 6.8k
Description
What happened + What you expected to happen
When I use ray 2.3, I found temp-dir can not work correctly in ray job submit mode.
ray job submit
and ray job log
command would not return logs to the stdout.
my analysis:
-
dashboard/modules/job/job_agent.py:get_job_logs -> dashboard/modules/job/job_manager.py:get_log_file_path -> python/ray/_private/node.py:get_logs_dir_path
-
the _log_dir is update by python/ray/_private/node.py:_init_temp
I think maybe commit e1a8796#diff-e486e78dca43662c19b202692ed44f45c7d49405a56b9ef58c56d85ab1ae2cc8 leads to this problem, because the self._ray_params.temp_dir will not be None to trigger internal_kv_get_with_retry to query the correct temp-dir
these problem also leads to ray client mode log would output to /tmp/ray, not the specific temp-dir
Versions / Dependencies
pip install ray==2.3
Reproduction script
// on physical node
ray start --head --block --disable-usage-stats --node-ip-address={NODE_IP} --block --dashboard-host='' --temp-dir=/opt/ray
// on you laptop
export RAY_ADDRESS=http://{NODE_IP}:8265
ray job submit --runtime-env-json='{"working_dir": "./"}' -- python3 counter.py
the logs would only output at the end of Job 'raysubmit_MqJfn5Ns8f2krGFQ' succeeded
ray job logs raysubmit_MqJfn5Ns8f2krGFQ is the same
Issue Severity
Medium: It is a significant difficulty but I can work around it.