Message-ID: <771956892.9007.1711713300680.JavaMail.confluence@host3.pipelinefx.com> Subject: Exported From Confluence MIME-Version: 1.0 Content-Type: multipart/related; boundary="----=_Part_9006_1951653310.1711713300680" ------=_Part_9006_1951653310.1711713300680 Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: quoted-printable Content-Location: file:///C:/exported.html
#################################################
@RELEASE: 6.5-3
#################################################
=3D=3D=3D=3D CL 12442 =3D=3D=3D=3D
@FIX: Fixed a corner-case MySQL permission problem with OSX/Linux superv= isor and the qube_readonly user.
Fixed by adding a "GRANT SELECT" with an explicit hostname (fe=
tched via
"SELECT @@hostname"), as in:
GRANT SELECT ON *.* TO 'qube_readonly'@'mysqlserverhostname'
JIRA: QUBE-438
=3D=3D=3D=3D CL 12358 =3D=3D=3D=3D
@FIX:Fixed example python script=
s so import of qb module will work in most cases.
=3D=3D=3D=3D CL 12347 =3D=3D=3D=3D
@FIX: pyCmd* jobtypes report all=
subsequent frames as failing when a 'regex_error' is matched and a frame i=
s marked as failed
=3D=3D=3D=3D CL 12339 =3D=3D=3D=3D
@FIX: fixed inaccurate worker ho=
st memory reporting on Windows platforms
ZD: 11367
=3D=3D=3D=3D CL 12333 =3D=3D=3D=3D
@FIX: worker shutdown code (QbWo=
rker::hostShutdown() and sendHostReport()) will now give up a lot quicker w=
hen being unable to contact the supervisor, instead of retrying for a long =
time.
=3D=3D=3D=3D CL 12322 =3D=3D=3D=3D
@FIX: issue where job instances =
don't terminate properly when very early kill/interrupt orders come in.
Sometimes interrupts and kills can come in before the worker has a chanc= e to properly complete the launching process of the proxy.exe process and i= ts main thread, causing unexpected behavior, such as a never-dying job inst= ance.
ZD: 11409
=3D=3D=3D=3D CL 12315 =3D=3D=3D=3D
@FIX: bug in initialization code=
of the QbJob class that messed up comparisons of jobs when sorting, which,=
among other things, prevented FIFO/FCFS ordering to be compromised. Now FI=
FO dispatching behavior should be more closely followed by jobs of equal pr=
iority (although not 100% strictly, due to the nature of the multithreaded =
architecture of the supervisor).
ZD: 11259
@INTERNAL TWEAK: added more debugging code to QbSupervisorQueue module.<= /p>
=3D=3D=3D=3D CL 12311 =3D=3D=3D=3D
@FIX: adding in ubuntu support: =
use bash explicitly rather than sh, specify 'awk' in location found on all =
OS's
=3D=3D=3D=3D CL 12306 =3D=3D=3D=3D
@FIX: issue where auto-expanded =
subjobs (instances) don't inherit the "retrysubjob" value set in =
the parent job, causing them NOT to auto-retry properly on failure.
ZD: 11292
=3D=3D=3D=3D CL 12298 =3D=3D=3D=3D
@FIX: Python API routines, such =
as qb.retrywork(), expecting workID as input would behave erroneously (such=
as retrying ALL agenda items on ALL jobs) when input a subjobID instead. V=
ice versa for routines expecting subjobIDs, such as qb.retry().
These<=
/p>
ZD: 11372
=3D=3D=3D=3D CL 12295 =3D=3D=3D=3D
@NEW: add support for new 'exiti=
ng' status
=3D=3D=3D=3D CL 12272 =3D=3D=3D=3D
@FIX: unreliable behavior when f=
requently modifying "cpus" of jobs up and down.
ZD: 11288
=3D=3D=3D=3D CL 12257 =3D=3D=3D=3D
@FIX: bug where auto-expand subj=
obs are incorrectly auto-retired, and in turn caused them NOT to expand any=
more.
ZD: 11217
=3D=3D=3D=3D CL 12255 =3D=3D=3D=3D
@FIX: issue where, if some inter=
mediate job processes crash and die unexpectedly, other job processes may b=
e missed by the cleanup code and left behind as zombies.
ZD: 11236
=3D=3D=3D=3D CL 12250 =3D=3D=3D=3D
@FIX: WorkerConfigFile makes a b=
etter effort at finding the worker config file, previously would save to de=
fault location when the file is actually in a non-default location as speci=
fied by supervisor_worker_configfile.
=3D=3D=3D=3D CL 12237 =3D=3D=3D=3D
@FIX: avoid inserting duplicated=
values into the 'outputPaths' for a frame when retried
=3D=3D=3D=3D CL 12232 =3D=3D=3D=3D
@FIX: "UnboundLocalError: l=
ocal variable 'qb' referenced before assignment" - issue experienced b=
y single customer on linux, re-importing qb module in main() resolves the i=
ssue. ZD# 11218
=3D=3D=3D=3D CL 12230 =3D=3D=3D=3D
@FIX: additional fixes to remedy=
"retrywork" issue with maya (and possibly other Perl-API based) =
jobs. See also the previous CL12228
=3D=3D=3D=3D CL 12228 =3D=3D=3D=3D
@FIX: automatic retry of agenda =
via "retrywork" not working properly in perl-based backends.
ZD: 11167
=3D=3D=3D=3D CL 12226 =3D=3D=3D=3D
@FIX:Fix issue where job_cleanup=
script would fail if run on a supervisor that did not have the MySQLdb pyt=
hon module installed.
=3D=3D=3D=3D CL 12219 =3D=3D=3D=3D
@FIX: "sre_constants.error:=
bogus escape (end of line)" - python-based jobs can crash on Windows =
at startup if path wrapped in QB_CONVERT_PATH() ends with a fwd-slash and h=
as being converted to a back-slash
=3D=3D=3D=3D CL 12177 =3D=3D=3D=3D
@FIX: Additional changes to supp=
ort proper Windows privilege enabling, added in CL12176
=3D=3D=3D=3D CL 12176 =3D=3D=3D=3D
@FIX: Add call to Windows' Adjus=
tTokenPrivileges() to explicitly enabled required privileges before launchi=
ng job instance (proxy) process
=3D=3D=3D=3D CL 12098 =3D=3D=3D=3D
@FIX: support negative frame ran=
ge in QB_* token parsing
=3D=3D=3D=3D CL 12082 =3D=3D=3D=3D
@FIX: issue where "modify&q=
uot;-ing the "cpus" value of a running job may incorrectly retire=
more instances than asked for.
This was due to race conditions of supe threads, and in extreme cases, w=
as
prematurely retire-ing ALL instances of a job while there are still=
pending
agendas, resulting in the job's instances to be all "com=
plete" but the job
itself to become "failed" since ther=
e are still pending agendas.
ZD: 10868
=3D=3D=3D=3D CL 12065 =3D=3D=3D=3D
@INTERNAL TWEAK: added/modified/=
corrected comments and symbol names for readability