13:37 <blaman> I'm running systemtap to tell me when a process hits the open file limit. Like so: stap -F -o /tmp/rlimit_nofile.log -S 1 /usr/share/doc/systemtap-client-3.0/examples/process/rlimit_nofile.stp. But then I get lines like this in the log? "Tue Aug 29 13:32:53 2017 python3.5(30867) open: 47 - max: 1024 Hit: EMFILE". Why does this trigger? It's saying the max is 1024, so I don't see why it would print that line when it only has 47 f

13:38 <blaman> stp script at https://sourceware.org/systemtap/examples/process/rlimit_nofile.stp

13:40 <fche> the error notification is not instant when the fd excess took place

13:41 <fche> it seems as though if the error condition hits a process once, it will be reported (repeatedly?) later, even if the number of open files becomes small again

13:45 <fche> might try this script (untested!) --> probe kernel.function("*alloc_fd").return { if ($return < 0) { t=task_current(); open_ds = task_open_file_handles(t); max_ds=task_rlimit_nofile(t) printf("%s(%d) alloc-fd fail (%d), cur=%d max=%d\n", execname(), pid(), $return, open_ds, max_ds) } }

13:46 <fche> basically squishing together the error detection and its reporting

14:01 <blaman> running, let's see..

14:11 <blaman> fche, seems to work. thanks!

14:11 <fche> woohoo

14:11 <fche> did it give sensible results?

14:12 <blaman> yep

14:12 <blaman> same numbers in cur and max as i'd expect

14:13 <fche> if you were inclined, you could submit that new script to replace the previous one, as a patch

14:14 <blaman> that'd require i fully understand how it works which is not the case right this moment :)

14:14 <fche> righto. just fixing the indentation may make it clearer :)

15:56 <slowfranklin> Hey folks!

15:57 <slowfranklin> Is there a way to avoid specifying absolute paths for SDT probes in libraries not installed in the default library search path dirs?

15:58 <slowfranklin> Eg I have some SDT probes in a library libsmbd-base-samba4.so located in

15:58 <slowfranklin> /home/slow/git/samba/scratch/bin/default/source3/libsmbd-base-samba4.so

15:59 <slowfranklin> and I'd like to avoid giving the full path in scripts.

15:59 <slowfranklin> I've tried setting LD_LIBRARY_PATH, not luck

16:00 <slowfranklin> Ie

16:00 <slowfranklin> probe process("libsmbd-base-samba4.so").provider("smb2").mark("pkt_recv_end")

16:00 <slowfranklin> fails

16:01 <slowfranklin> , but

16:01 <slowfranklin> probe process("/home/slow/git/samba/scratch/bin/default/source3/libsmbd-base-samba4.so").provider("smb2").mark("pkt_recv_start")

16:01 <slowfranklin> works

16:03 <irker016> systemtap: dsmith systemtap.git:refs/heads/master * release-3.1-248-gcd8837d / runtime/linux/copy.c runtime/linux/loc2c-runtime.h: Fix PR22012 by updating the way we read user strings. http://tinyurl.com/yc44flxn

16:53 <drsmith> slowfranklin: one way to handle that would be use command line args, like: stap -ve 'probe process(@1).provider("smb2").mark("pkt_recv_start") { /* code here */ }' home/slow/git/samba/scratch/bin/default/source3/libsmbd-base-samba4.so

16:54 <slowfranklin> Hm. Damn. :)

16:54 <slowfranklin> I'm already thinking about using environment variables. I'm embedding the stap script in a bash script anyway...

16:56 <slowfranklin> Is it actually correct that I can specify just the library name as long as it's in a standard path, say /lib64?

16:56 <slowfranklin> Or do I alway have to specify the full path for libraries?

16:56 <slowfranklin> Seems a bit awkward...

16:57 <drsmith> Good question.

16:57 <slowfranklin> Parts of the documentation suggests * works in a few places, but it doesn't work here.

16:57 <drsmith> I'm not sure this helps you, but you can use wildcards in those paths: probe process("/lib*/libpthread.so").function...

16:58 <slowfranklin> That would help if it would work... :)

16:58 <drsmith> Wildcards should work in library paths, assuming your version of systemtap supports them.

16:58 <slowfranklin> $ stap -V

16:58 <slowfranklin> Systemtap translator/driver (version 3.1/0.169, rpm 3.1-5.fc26)

16:58 <slowfranklin> Fedora 26 default package

16:59 <drsmith> oh definitely supported there

16:59 <slowfranklin> Throws strange errors, hold on...

17:00 <slowfranklin> Urks: https://cpaste.org/?1b90740a0253d909#xRLMXyfs7UJQmJPjmLEtnrkhDg9F5kvzzTZEc5q9yzQ=

17:01 <drsmith> that is odd

17:02 <drsmith> were you expecting that wildcard to take the place of > 1 directory?

17:03 <slowfranklin> sure, yes please

17:03 <drsmith> nope, they only work for 1 level

17:03 <slowfranklin> *ouch*

17:03 <drsmith> think of them as a bash-style wildcard

17:03 <slowfranklin> Well, the path to the library is /home/slow/git/samba/scratch/bin/default/source3/libsmbd-base-samba4.so

17:03 <slowfranklin> so...

17:04 <slowfranklin> But that's just the path in my development non-install build

17:04 <drsmith> right

17:04 <slowfranklin> Configured prefix is /opt/samba/

17:04 <slowfranklin> So another path

17:04 <slowfranklin> Depending on the configure paths I'd might be anywhere

17:04 <drsmith> if you are already embedding the stap script in a bash script, seems like you could just treat the stap script as a string and do some bash variable substitution on it

17:05 <slowfranklin> And I'd like to cater for that in the script so that it somehow works out of the box

17:05 <slowfranklin> yeah, that's what I'm now working on as a workaround

17:06 <slowfranklin> I'll use some sane default like /lib64 and allow overriding by passing env variables.

17:06 <slowfranklin> At least now I know it's not pebkac :)

17:07 <drsmith> you might want the default to be something like "/lib*/libsmbd-base-samba4.so", so that handles /lib64 and/or /lib

17:07 <slowfranklin> yup. Thanks!

20:26 <irker016> systemtap: dsmith systemtap.git:refs/heads/master * release-3.1-249-g9d083c7 / dwflpp.cxx: Fix PR22031 by initializing the context class instance before using it. http://tinyurl.com/y9q8964h

21:27 <irker016> systemtap: dsmith systemtap.git:refs/heads/master * release-3.1-250-gf630ccf / buildrun.cxx runtime/linux/autoconf-bio-bi_opf.c tapset/linux/ioblock.stp: Fix PR22036 by handling "struct bio" kernel changes. http://tinyurl.com/ybtmr7no