|
18 | 18 |
|
19 | 19 | *** 24.12.14 bugs |
20 | 20 |
|
21 | | -- Fixed a bug that caused the Job Router to abort after an upgrade from |
22 | | - version 24.0, if the job queue contained jobs submitted by a client |
23 | | - running 24.X or later. |
| 21 | +- Fix interoperability problem between HTCondor-CE 24 and 25 which |
| 22 | + manifests as a Job Router crash when upgrading the CE to HTCondor 25 |
24 | 23 | :jira:`3355` |
25 | 24 |
|
| 25 | +- Fixed a bug in :meth:`Schedd.submit` where single-entry itemdata could be |
| 26 | + truncated to its first character. |
| 27 | + :jira:`3272` |
| 28 | + |
| 29 | +- Fixed a bug in :meth:`Schedd.submit` where the ``$(step)`` submit variable |
| 30 | + wasn't being set. |
| 31 | + :jira:`3272` |
| 32 | + |
| 33 | +- Fixed a bug in :meth:`Submit.itemdata` causing multi-entry itemdata to be |
| 34 | + returned incorrectly. |
| 35 | + :jira:`3272` |
| 36 | + |
26 | 37 | - Fixed a bug where using :subcom:`max_idle`, :subcom:`container_image` |
27 | 38 | and :subcom:`transfer_input_files` could result in the container only |
28 | 39 | being transferred along with the first job. |
|
39 | 50 | :class:`classad2.ExprTree`, preventing segmentation faults when using the object. |
40 | 51 | :jira:`3319` |
41 | 52 |
|
42 | | -- Fixed a bug in :meth:`Schedd.submit` where single-entry itemdata could be |
43 | | - truncated to its first character. |
44 | | - :jira:`3272` |
45 | | - |
46 | | -- Fixed a bug in :meth:`Schedd.submit` where the ``$(step)`` submit variable |
47 | | - wasn't being set. |
48 | | - :jira:`3272` |
49 | | - |
50 | | -- Fixed a bug in :meth:`Submit.itemdata` causing multi-entry itemdata to be |
51 | | - returned incorrectly. |
52 | | - :jira:`3272` |
53 | | - |
54 | 53 | - Fixed a bug where :tool:`condor_history` could take many minutes to read a single line |
55 | 54 | from the history file when the line is many megabytes long. This was causing |
56 | 55 | ingestion of epoch ads into a database to timeout. |
|
73 | 72 |
|
74 | 73 | *** 24.0.14 bugs |
75 | 74 |
|
76 | | -- To work around issues with updated CUDA runtime, when the *condor_starter* |
77 | | - sets the environment variable ``NVIDIA_VISIBLE_DEVICES`` for a job, it now |
78 | | - forces ``CUDA_VISIBLE_DEVICES`` to have the same value. The values will |
79 | | - always have the long form GPU ids if any GPUs are assigned to the job. |
| 75 | +- Fix problem running PyTorch jobs on multiple GPUs with |
| 76 | + newer versions of the CUDA library by providing long GPU IDs |
| 77 | + in the ``CUDA_VISIBLE_DEVICES`` environment variable |
80 | 78 | :jira:`3350` |
81 | 79 |
|
82 | 80 | - Calling :meth:`htcondor2.Schedd.history` on an object whose corresponding |
|
0 commit comments