| To: | PCP <pcp@xxxxxxxxxxx> |
|---|---|
| Subject: | problem with last change to _get_pids_by_name() in pcp.env |
| From: | Ken McDonell <kenj@xxxxxxxxxxxxxxxx> |
| Date: | Wed, 21 Sep 2016 09:16:56 +1000 |
| Delivered-to: | pcp@xxxxxxxxxxx |
| User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.2.0 |
Commit 63ce703 forces matching to the full pathname of the binary (not a
bad idea in principle, and motivated by a real issue) ... but the
implementation fails (spectacularly) when the
(a) the shell running _get_pids_by_name() has a different $PATH than the shell that was used to launch the binary in question, so which(1) and exec(2) may refer to different binaries, and/or (b) the binary in question was launched via a symbolic link.Specifically pmlogger (on at least some distros) lives in /usr/bin and as a symlink in $PCP_BINADM_DIR ... I have a bunch of QA failures from the resulting mismatch. I don't have a fix as yet, but this is a heads up for others pending resolution. |
| <Prev in Thread] | Current Thread | [Next in Thread> |
|---|---|---|
| ||
| Previous by Date: | [Bug 1376856] pmlogger pmcd-restart persistence breaks pmmgr assumptions, bugzilla |
|---|---|
| Next by Date: | pcp updates: important code fixes and qa, Ken McDonell |
| Previous by Thread: | [Bug 1377464] New: "pmrep proc" takes too long, bugzilla |
| Next by Thread: | pcp updates: important code fixes and qa, Ken McDonell |
| Indexes: | [Date] [Thread] [Top] [All Lists] |