Change 147 on 2000/03/19 by S.Herbert@sheffield.ac.uk Update release date info for v3.50.9 Change 146 on 2000/03/19 by S.Herbert@sheffield.ac.uk Change version numbers to reflect new GNQS release. Change 145 on 2000/03/19 by S.Herbert@sheffield.ac.uk Generic NQS v3.50.9 created Change 144 on 2000/03/16 by andrews@iname.com Remove a error message from the users console by reducing the severity of the message. Requested by Christoph Schnorrenberger <schnorre@rhrk.uni-kl.de> to NQS-Support on 2000-03-15. Change 141 on 2000/03/07 by andrews@iname.com Fix problem with NFS and root permissions. Change 140 on 2000/03/05 by andrews@iname.com Since so much code relies on mid=0 to be the local host, we prevent a user from creating a global mid=0. Change 138 on 2000/03/02 by andrews@iname.com Migrate dynamic environment size to Stable release. Eliminates two hard coded limits (MAX_ENVIRONMENT and MAX_ENVIRONVARS) and allows long lines in control file. Change 131 on 2000/02/21 by S.Herbert@sheffield.ac.uk Check-in of remaining GSetup files before we remove the 'all-systems' directories. Change 130 on 2000/02/17 by andrews@iname.com Fix declaration for Locmid in delreq.c Change 129 on 2000/02/17 by andrews@iname.com Implement the errno fix in SIGCHLD handler. Take ideas from netdaemon that everything manages their child exits. Change 128 on 2000/02/17 by andrews@iname.com Fix times printed in syslog. The current system prints GMT/UTC without labeling it as such. Changed to use localtime. Change 127 on 2000/02/17 by andrews@iname.com Change SIGCHLD handling in netdaemon. Hopefully eliminate a problem with setting errno in reapchild and "tricking" other error checks into thinking they failed. Change 125 on 2000/02/14 by S.Herbert@sheffield.ac.uk Snapfiles now correctly list the name of the user who created the snapfile in the comment block at the top of the snapfile. Change 124 on 2000/02/12 by andrews@iname.com Fix scheduler bug if servers reporting their load averages to the scheduler do not have a server performance index in the scheduler database. Reported by: Gotthard Saghi-Szabo <gotthard.saghi-szabo@gl.ciw.edu> Change 123 on 2000/02/11 by S.Herbert@sheffield.ac.uk Complete re-working of how NQSdaemon decides to check its queues. NQSdaemon no longer uses UNIX alarms to schedule tasks such as checking queues for jobs. This can now be done inside the main nqs_main() loop instead. This required the following: - getpchar() in nqs_main.c will now timeout after 5 secs - interread() now supports reads that timeout - nqs_main() now supports read timeouts - nqs_main() now calls nqs_valarm() everytime it goes round the main loop - nqs_valarm() no longer installs a signal handler - setalarm() in nqs_vtimer.c no longer sets alarms - more debugging messages added This change (on my setup here) means that pipeclient no longer hangs when trying to talk to its NQS shepherd process. This in turn removes another way in which jobs get stuck in the arriving state. This fix *may* (I cannot test it here) also fix the signal problems plaguing some DEC Alpha users. Once this fix has been tested, and verified to work for others, there is more to be done: - examine and optimise nqs_main() loop - examine and verify that nqs_valarm() cleans up the alarm queue correctly - examine and optimise nqs_valarm() and nqs_vtimer() *** WARNINGS *** - running NQS with the debugging level set to 4 or higher will now fill your logfiles pretty quickly! Run with debugging set to 1 unless you're creating a logfile to help us examine a bug - NQSdaemon will now use more CPU time than previous versions did. This should not be too much of a problem, and future optimisation of the timer code will probably claim most of it back. Change 122 on 2000/02/11 by S.Herbert@sheffield.ac.uk Slight change to the 'jobs stuck in arriving state' fix. Should work this time ;-) And the qmgr command is 'qmgr set default destination_retry wait'. Change 121 on 2000/02/11 by S.Herbert@sheffield.ac.uk Another fix for jobs stuck in arriving state bug. This time, I've made sure that NQSdaemon's nqs_quereq() will check for, and prevent two requests with the same seqno and mid existing at the same time. This means that a request will be deferred, and will be retried at a later time. The delay on a retry is normally 5 mins; use 'qmgr set default destination_retry time' to change this to 30 seconds or something. Change 120 on 2000/02/10 by S.Herbert@sheffield.ac.uk Added Changes documents from v3.50.8 to Current tree Change 119 on 2000/02/10 by S.Herbert@sheffield.ac.uk GNQS_Stable/Current/ now set to version v3.50.9 Change 118 on 2000/02/10 by andrews@iname.com Minor bug fix. Include the '=' character in the security check comparision. This prevents a user setting the variable ENV1=value from matching ENV11=value. Change 117 on 2000/02/09 by S.Herbert@sheffield.ac.uk Added version-specific changelogs, and updated the release date, prior to release.