thanxs, upgrading to 3.2.2 did the trick! now the workers are working. also thanxs for the config options, we have used a configure-script which we have created with long time ago with squid 2.5x and then adjusted it a little. some of the options have survived since then... mfg Markus Rietzler <rietzler_software/> Rechenzentrum der Finanzverwaltung Tel: 0211/4572-2130 > -----Ursprüngliche Nachricht----- > Von: Amos Jeffries [mailto:squid3@xxxxxxxxxxxxx] > Gesendet: Mittwoch, 10. Oktober 2012 23:59 > An: squid-users@xxxxxxxxxxxxxxx > Betreff: Re: squid 3.2.1: workers not working > > On 11.10.2012 03:42, Rietzler, Markus (RZF, SG 324 / > <RIETZLER_SOFTWARE>) wrote: > > we are trying to get squid using SMP workers. tried with workers2 > > seting up different cache_dirs for each worker > > > > workers 2 > > http_port 8080 > > cache_dir aufs $SQUID_CACHE_ROOT/${process_number} 32000 16 256 > > > > Seems okay. But note the 32GB cache size. > > > when we start squid we get: > > > > 2012/10/10 16:26:16.663 kid3| IoCallback.cc(107) finish: called for > > local=[::] remote=[::] FD 13 flags=1 (0, 0) > > 2012/10/10 16:26:16.663 kid1| fd_open() FD 19 > > /rzf/db/www/squid/1/swap.state.clean > > 2012/10/10 16:26:16.663 kid3| comm_read_try: FD 11, size 4328, retval > > 4112, errno 0 > > 2012/10/10 16:26:16.663 kid1| storeDirWriteCleanLogs: opened > > /rzf/db/www/squid/1/swap.state.clean, FD 19 > > 2012/10/10 16:26:16.663 kid3| IoCallback.cc(107) finish: called for > > local=[::] remote=[::] FD 11 flags=1 (0, 0) > > 2012/10/10 16:26:16.663 kid3| comm_close: start closing FD 13 > > 2012/10/10 16:26:16.663 kid3| comm.cc(735) commUnsetFdTimeout: Remove > > timeout for FD 13 > > 2012/10/10 16:26:16.663 kid1| fd_close FD 15 > > /rzf/db/www/squid/1/swap.state > > 2012/10/10 16:26:16.663 kid3| The AsyncCall comm_close_complete > > constructed, this=0x9e88e0 [call144] > > 2012/10/10 16:26:16.663 kid1| Cache Dir #0 log closed on FD 15 > > 2012/10/10 16:26:16.663 kid3| comm.cc(1154) will call > > comm_close_complete(FD 13) [call144] > > 2012/10/10 16:26:16.663 kid1| xrename: renaming > > /rzf/db/www/squid/1/swap.state.clean to > > /rzf/db/www/squid/1/swap.state > > 2012/10/10 16:26:16.663 kid3| Coordinator.cc(146) > > handleSharedListenRequest: kid1 needs shared listen FD for > > 130.11.6.5:8080 > > 2012/10/10 16:26:16.664 kid3| Coordinator.cc(154) > > handleSharedListenRequest: sending shared listen > > local=130.11.6.5:8080 > > remote=[::] FD 15 flags=9 for 130.11.6.5:8080 to kid1 mapId=0 > > 2012/10/10 16:26:16.664 kid3| entering comm_close_complete(FD 13) > > 2012/10/10 16:26:16.664 kid3| AsyncCall.cc(34) make: make call > > comm_close_complete [call144] > > 2012/10/10 16:26:16.664 kid1| fd_open() FD 15 > > /rzf/db/www/squid/1/swap.state.last-clean > > 2012/10/10 16:26:16.664 kid3| fd_close FD 13 > > 2012/10/10 16:26:16.664 kid1| fd_close FD 15 > > /rzf/db/www/squid/1/swap.state.last-clean > > 2012/10/10 16:26:16.664 kid3| leaving comm_close_complete(FD 13) > > 2012/10/10 16:26:16.664 kid1| fd_close FD 19 > > /rzf/db/www/squid/1/swap.state.clean > > 2012/10/10 16:26:16.664 kid3| comm.cc(2116) comm_open_uds: Attempt > > open socket for: /rzf/produkte/www/squid/var/run/squid/kid-1.ipc > > > 2012/10/10 16:26:16.664 kid1| Finished. Wrote 0 entries. > > 2012/10/10 16:26:16.664 kid1| Took 0.00 seconds ( 0.00 > > entries/sec). > > 2012/10/10 16:26:16.664 kid3| comm.cc(2134) comm_open_uds: Opened UDS > > FD 13 : family=1, type=2, protocol=0 > > 2012/10/10 16:26:16.664 kid3| fd_open() FD 13 > > 2012/10/10 16:26:16.664 kid3| comm.cc(748) commSetConnTimeout: > > local=[::] remote=[::] FD 13 flags=1 timeout 10 > > FATAL: kid1 registration timed out > > FATAL: kid2 registration timed out > > > > and later: > > > > 2012/10/10 16:26:19.675 kid2| leave_suid: PID 11031 called > > 2012/10/10 16:26:19.675 kid2| leave_suid: PID 11031 giving up root, > > becoming 'www' > > 2012/10/10 16:26:19.675 kid2| leave_suid: PID 11031 called > > FATAL: Ipc::Mem::Segment::open failed to > > shm_open(/squid-squid-page-pool.shm): (2) No such file or directory > > > > This is kid2 attempting to open its UDS connection to register with the > coordinator (kid3). > > > > but I think this comes from the first FATAL. > > It's related to the "FATAL: kid2" but this one is caused by the UDS > path not existing. The shm_open() is supposed to create one if it does > not exist already. > > Start with an upgrade to 3.2.2 we fixed some SHM related bugs there. > Then check bugzilla for more info, shm_open() has a few OS-specific > problems and oath problems known. > > > > > > why are the kids nor registered. squid (coordinator) is running but > > no other squid process and so it does not listen on port 8080. > > > > The registration is done by sending packets to their UDS sockets (the > SHM path which failing to open). > > To put it in more familiar terms; what is happening is somewhat vaguely > equivalent to a network socket() creation failing to happen in the > worker, which prevents connect() being done and the server/coordinator > waiting for a connect() SYN packet never receives one. > (Bit more complex than that, but essentially similar.) > > > > > how can we adjust the path where squid stores the ipc files for the > > coordinator and kids? it is /path/to/squid/var/run/squid. this is > > because we used -prefix="/path/to/squid" . but also we have used > > -prefix we want to have the files in other dirs. with (nearly) all we > > can adjust the locations in squid.conf and use the full/other path, > > The OS determines where/what the SHM path descriptor has to be. We like > to follow the FHS specification since these are special networking > *socket* descriptors not "files". That may or may not permit your > --prefix to apply on the path, but we do not allow localization. > > > > > > here the output of squid -v with compile opitons. do we have to set > > one option to have workers workering? > > No. SMP has no ./configure options. It is always built when supported > by the system. The "workers" config file directive is how you turn it > on/off (off being 'workers 1' or absent from the file). > > > > > Squid Cache: Version 3.2.1 > > configure options: '--enable-basic-auth-modules=MSNT,SMB,NEGOTIATE' > > --enable-basic-auth-modules does not exist. never has AFAICT. > > I assume you mean: --enable-auth-basic="MSNT SMB NEGOTIATE" > > * I am also assuming you have a local basic auth module called > "NEGOTIATE" which you are patching into the Squid sources, since we > don't publish any such helper. > > > > '--enable-external-acl-helpers=ldap_group' > > '--enable-auth-basic' > > see above. > > > > '--enable-auth-ntlm' > > '--enable-auth-negotiate=squid_kerb_auth' > > --enable-auth-negotiate=kerberos > > otherwise the negotiate_kerberos_auth binary will not build. > > > > '--enable-ntlm-fail-open' > > --enable-ntlm-fail-open is obsolete. has not worked since *before* > squid-2.5. > > > '--enable-delay-pools' > > '--enable-follow-x-forwarded-for' > > '--with-maxfd=4095' > > --with-maxfd was an experiment by RHEL. The squid option is > --with-filedescriptors=4095 > > > > '--enable-removal-policies=lru,heap' > > '--with-winbind' > > --with-winbind does not exist. > > > '--with-async-io' > > '--enable-storeio=ufs,aufs' > > '--disable-ident-lookups' > > '--prefix=/path/to/squid' > > '--enable-err-language=German' > > --enable-err-language is obsolete. > > > '--enable-underscores' > > '--with-large-files' > > '--enable-dlmalloc' > > --enable-ltdl-convenience > > > > mfg > > > > Markus Rietzler > > <rietzler_software/> > > Rechenzentrum der Finanzverwaltung > > > > Tel: 0211/4572-2130
