Ticket #1001: Default WANT_SUSPEND policy goes UNDEFINED, startd EXCEPTs
Activation timer showed up in 1998 (git show 3cef52db). The conditional setting of ATTR_JOB_START has been around since at least 2003 (see V6_5-branch).
See: https://bugzilla.redhat.com/show_bug.cgi?id=540545
Remarks:
2009-Nov-25 17:17:30 by tannenba:
So... bold proposal: all startd policy expressions must meta-equals True to be triggered. In other words, if any of the policy expressions evaluate to UNDEFINED it is the same as evaluating to False. Thoughts?
2009-Nov-30 08:55:38 by tstclair:
The code actually did the "right" thing, but it was rather difficult to diagnose. I believe this points to a more fundamental issue, leaving breadcrumbs in error handling. The original error was EXCEPT( "Can't find WANT_SUSPEND in internal ClassAd" ), which gives a starting point, but developers have to whip out their sloothing hats and determine a direction.
It may be beneficial to examine the EXCEPT sites and leave breadcrumbs (e.g. - "Hint: Check that your expression does not have an UNDEFINED variable.")
maybe an EXCEPT_TYPE(T, "Message")
T appends debugging info based on type.
Check a, then b, then c, if all else fails nuke this code.
To be honest, I would use exceptions, or a tribool... but hey, that's me. ;-)
2009-Nov-30 09:36:52 by matt:
I would be for pulling the policy expressions into a central place where we can document and enforce consistency.
Properties:
| Type: |
defect |
|
Last Change: |
2009-Dec-18 15:02 |
| Status: |
resolved |
|
Created: |
2009-Nov-25 11:50 |
| Fixed Version: |
v070402 |
|
Broken Version: |
v070000 |
| Priority: |
3 |
|
Subsystem: |
Daemons |
| Assigned To: |
matt |
|
Derived From: |
|
| Creator: |
matt |
|
Rust: |
|
| Customer Group: |
other |
|
Visibility: |
public |
| Notify: |
matt@cs.wisc.edu,zmiller@cs.wisc.edu,tannenba@cs.wisc.edu |
|
Due Date: |
|
Related Check-ins:
| 2022-Nov-28 14:25 |
|
Check-in [70706]: Merged [70695], Merge pull request #1001 from htcondor/V10_2-HTCONDOR-1337-no-sort-by-qdate HTCONDOR-1337-no-sort-by-qdate Committer: GitHub (By GregThain ) |
| 2010-Nov-29 15:35 |
|
Check-in [25797]: more correctly document WANT_SUSPEND behavior -- changed in 7.4.2 (gittrac #1001) (By Karen Miller ) |
| 2010-Nov-29 15:05 |
|
Check-in [25795]: Revert "Documented meaning of UNDEFINED WANT_SUSPEND. #1001" This reverts commit ba7dae13e5d9103838cd5bcf5753410c03f83fda. (By Karen Miller ) |
| 2010-Nov-29 12:26 |
|
Check-in [25794]: Documented meaning of UNDEFINED WANT_SUSPEND. #1001 (By Dan Bradley ) |
| 2009-Dec-18 14:55 |
|
Check-in [16728]: WANT_SUSPEND that is UNDEFINED, either by being absent or evaluating to UNDEFINED, now means FALSE. It used to mean EXCEPT! (#1001) (By Matthew Farrellee ) |
| 2009-Dec-18 14:50 |
|
Check-in [16727]: Protect ActivationTimer from going to UNDEFINED (#1001) (By Matthew Farrellee ) |