potiuk commented on PR #35181: URL: https://github.com/apache/airflow/pull/35181#issuecomment-1788770992
> This doesn't flush the existing metrics when you create a new instance? For example, graphs still have their data from before the flush?? I believe there are virtually no metrics before. Daemonisation happens right at the very start of any component that enables it, and I think what this problem mainly solves is that it seems Statsd integration when imported and initialized saves some states and possibly starts some thread that then are "lost" when daemonizing. The deamonisation does a few things: 1) forks the process (twice I believe) to make sure that it detaches from the process that started it and moves itself to have "init" process as parent so that any signals (SIGTERM/SIGHUP) from the original terminal process do not kill the background process 2) closes stdin/stdout and opened files to make sure it does not continue writng / reading from terminal which was used to start it and that the files are not accessed from multiple processes at once. I think what happens here - is that Statsd manages to open a socket / file descriptor before daemonization happnes and that socket gets closed when forking/daemonising. When deamonization happens, the parent/intermediate processes exit right after the fork happens. So even if Statsd managed to open the socket before daemonization - it gets closed right after (and by reinitialising Statsd in this PR we are reopening it in the forked process) But also writing this comment makes me think: If we manage to find out the resource (likely socket) that Statsd opens and get hang of file descriptor for that - there is a better way of handling it https://daemonize.readthedocs.io/en/latest/ - you can add such file descriptor to "keep_fds" list of descriptors that should NOT be closed when forking and their ownership passed to the forked process. So @pavansharma36 - maybe you can investigate what Statsd does under the hood to make it stop working - find the descriptor it opens at initialization and pass it in keep_fds parameter? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: commits-unsubscr...@airflow.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
