Bug 523806

Created attachment 361766 [details]
Patch for spurious spool directory creation and deletion

This patch prevents the schedd from creating a job sandbox when running jobs without file transfer (as determined by "should_transfer_files = NO" in the job ad).

This patch is built in 7.4.0-0.5.

Tested with submission then removal of 100 jobs. The -0.4 data may have let the jobs run to completion. Data gathered with strace -c -p `pidof condor_schedd`.

$ grep -e mkdir -e fsync -e stat -e open -e chown -e rmdir nfs-* local-*

nfs-0.4: 21.92    0.012985          65       200           mkdir
nfs-0.4: 20.23    0.011987          40       301           fsync
nfs-0.4: 10.01    0.005930           1      4122       300 stat
nfs-0.4:  8.83    0.005230           1      4576      1884 open
nfs-0.4:  8.43    0.004993          17       300           chown
nfs-0.4:  6.74    0.003994          20       200           rmdir
nfs-0.4:  0.09    0.000056           0      2695           fstat
nfs-0.4:  0.07    0.000040           0       800           lstat

nfs-0.5: 39.20    0.001998           6       308           fstat
nfs-0.5: 39.18    0.001997          18       114           fsync
nfs-0.5: 12.14    0.000619           1       457       300 stat
nfs-0.5:  0.00    0.000000           0        30         1 open

local-0.4: 12.21    0.001997           7       301           fsync
local-0.4:  5.66    0.000926           5       200           mkdir
local-0.4:  1.78    0.000291           0      3365      1351 open
local-0.4:  1.74    0.000285           0      3525       300 stat
local-0.4:  1.52    0.000249           1       200           rmdir
local-0.4:  0.42    0.000068           0       800           lstat
local-0.4:  0.31    0.000050           0       300           chown
local-0.4:  0.30    0.000049           0      2158           fstat

local-0.5: 84.42    0.002997          29       102           fsync
local-0.5:  2.08    0.000074           0       328       300 stat
local-0.5:  0.39    0.000014           1        20         1 open
local-0.5:  0.00    0.000000           0       318           fstat

Is there an upgrade issue with this patch? Will existing jobs that have spool directories and are not using file transfer have those directories leaked?

In that case, yes, those directories will leak.  For now, keep this in mind when upgrading running pools; I'll find a workaround.

*** Bug 523801 has been marked as a duplicate of this bug. ***

*** Bug 524644 has been marked as a duplicate of this bug. ***

The fix is in 7.4.0-0.5, and should be going upstream as well pending code review.

Release note added. If any revisions are required, please set the 
"requires_release_notes" flag to "?" and edit the "Release Notes" field accordingly.
All revisions will be proofread by the Engineering Content Services team.

New Contents:
In cases that require no file transfer (e.g. in which the spool directory is on a shared filesystem or the local filesystem), the condor schedd will no longer delete the spool directory when the job completes or is removed (523806)

Release note updated. If any revisions are required, please set the 
"requires_release_notes"  flag to "?" and edit the "Release Notes" field accordingly.
All revisions will be proofread by the Engineering Content Services team.

Diffed Contents:
@@ -1 +1 @@
-In cases that require no file transfer (e.g. in which the spool directory is on a shared filesystem or the local filesystem), the condor schedd will no longer delete the spool directory when the job completes or is removed (523806)+In cases that require no file transfer (e.g. in which the spool directory is on a shared filesystem or the local filesystem), the condor schedd will no longer maintain a job spool directory (523806)

Spool directories are not created anymore when they are not needed. Tested with the strace, both on local filesystem and a with remote (nfs) filesystem, on RHEL 4.8/5.4, i386/x86_64, condor-7.4.1-0.5.

Changing the status to VERIFIED.

Release note updated. If any revisions are required, please set the 
"requires_release_notes"  flag to "?" and edit the "Release Notes" field accordingly.
All revisions will be proofread by the Engineering Content Services team.

Diffed Contents:
@@ -1 +1,8 @@
-In cases that require no file transfer (e.g. in which the spool directory is on a shared filesystem or the local filesystem), the condor schedd will no longer maintain a job spool directory (523806)+Grid bug fix
+
+C: In cases where no file transfer is required, the condor schedd will attempt to stat, chown, and delete the spool directory when the job completes or is removed.
+C: In environments using NFS spool directories, these operations run extremely slowly, and may also overload the NFS server.
+F: The schedd was changed, and no longer creates or deletes job spool directories.  
+R: The bottleneck no longer occurs.
+
+In cases where no file transfer is required, the condor schedd will attempt to stat, chown, and delete the spool directory when the job completes or is removed. In environments using NFS spool directories, these operations run extremely slowly, and may also overload the NFS server. The schedd was changed, and no longer creates or deletes job spool directories, and the bottleneck no longer occurs.

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on therefore solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHEA-2009-1633.html