In the case of an MPI application running on a cluster, heaptrack by default creates many output files, each with pid attached to the output file name. This however is not enough to identify which output file corresponds to which pid runs on which machine. Would it be possible to let user specify something like "heaptrack -o my_preferred_output_%h_%p.zst", where %h indicates the hostname of the pid runs on, and %p means the pid ?
yes that sounds like a good idea. in general improving the ability to define the file name would be welcome. are you willing to provide a patch? It should be fairly easy to do. see existing code in src/track/heaptrack.sh.cmake which handles the `-o` switch.
Hi Milian, Sure, I can do the pull request. Do you want the default output file name being suffix'ed with hostname as well, for example do the following change? - output=$(pwd)/heaptrack.$(basename "$client").$$ + output=$(pwd)/heaptrack.$(basename "$client").$(hostname).$$
no I'd like to keep the current format and only allow %h and %p and similar as place holders in the `-o` parameter and then they should get replaced by the hostname, pid or whatever else we may want to add in the future.
Git commit 22e7a9cd9c028d7da9aaabad0b691f1b56f325d3 by Milian Wolff, on behalf of Jacob King. Committed on 07/04/2021 at 07:16. Pushed by mwolff into branch 'master'. feat: output file name under -o can include %h and %g M +11 -1 src/track/heaptrack.sh.cmake https://invent.kde.org/sdk/heaptrack/commit/22e7a9cd9c028d7da9aaabad0b691f1b56f325d3