-
Notifications
You must be signed in to change notification settings - Fork 5
/
RELEASENOTES.txt
101 lines (77 loc) · 3.58 KB
/
RELEASENOTES.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
Condor Agent Release Notes
==========================
Release 1.27
------------
Added:
* Scratch directory cleanup can be disabled by setting the HTCondor configuration attribute
CONDOR_AGENT_SKIP_CLEANUP to True
Fixed:
* Issue #1: Fixed a bug where a scratch directory could get deleted inadvertently if the clusterID
was not properly written to the .submit file.
Release 1.26
------------
Added:
<nothing>
Fixed:
* Case #13400: Fixed a bug introduced in 1.25 where condor_agent.py would try to register a handler
for SIGQUIT on Windows. Windows does not support that signal, so now only non-Windows operating
systems will attempt to handle SIGQUIT.
Release 1.25
------------
Added:
<nothing>
Fixed:
* Case #5157: The bash script wrapper for condor_agent.py no longer blocks SIGQUIT, which
prevents the agent from stopping HTCondor shutdown.
* Case #5416: HTCondor expects the HISTORY value to be a file, not a directory, so the agent will
now only check files for job history. This prevents performance impacts when HISTORY is
accidentally defined to be the same as LOG (or other directory with numerous/large files).
Release 1.21
------------
Fixed:
* No longer uses /etc/condor/config as a default for Windows.
Release 1.19
------------
Added:
<nothing>
Fixed:
* Case #8236: Race condition in cleanup on slow Condor startup
An overly burdened scheduler can take a while to return jobs when it first
starts up, despite actually having jobs in the queue. The cleanup thread needs
to sleep when it first starts up so it doesn't query a scheduler in this state,
think it's empty, and wipe all the temporary directories of the jobs that are
in the scheduler's queue.
Release 1.18
------------
Added:
* detailed documentation for accessing the REST API for submissions with a nice example
that uses curl from the command line to do a "remote" submission via the API.
Fixed:
* Remove a race condition between submissions and the cleanup thread
It's possible that the submission can complete but a condor_q call for the cluster doesn't
show any jobs in the queue yet. Busy schedulers can get behind in their processing. This
change adds a pickle file mtime consideration. The cleanup thread will skip inspecting
clusters if their pickled data on disk is less than 5 minutes old. That's a safe buffer.
* Case 7403: Submissions fail after a period of time
Tracked the problem down to using os.getcwd() -- this call is not thread-safe. It queries a
process-level global. This means a thread doing an os.chdir() can change the value that
*another* thread gets when it call os.getcwd().
Release 1.15
------------
Added:
* Committed a new RELEASENOTES.txt document to the repository, included in final packages
Fixed:
* Drop the mglob dependency
It was annoying and not overly useful. Replaced it with a glob+isfile check that does just
as well.
* Case 7355: Encode the scheduler name in the cluster file on disk to avoid file clashes
Without the scheduler name included in the file name it's possible that duplicate cluster
IDs will occur for a multi-scheduler machine which would would mean proxy submission
.cluster files would clash.
* Omit a \n when returning the cluster ID for a successful submission
The body of the response on success should just be the cluster ID, no newline
character required.
* Ensure submission cleaner thread doesn't die
On startup the thread checks for CONDOR_AGENT_SUBMIT_DIR -- if this isn't defined
the Agent shouldn't let the cleaner thread exit. Instead it should just warn that
it's not doing cleanup at this point in time and sleep as it usually does.