[Swift-commit] r2547 - trunk/tests/misc

noreply at svn.ci.uchicago.edu noreply at svn.ci.uchicago.edu
Sun Feb 15 18:04:35 CST 2009


Author: benc
Date: 2009-02-15 18:04:35 -0600 (Sun, 15 Feb 2009)
New Revision: 2547

Modified:
   trunk/tests/misc/clusters.sh
Log:
fix cluster test

Modified: trunk/tests/misc/clusters.sh
===================================================================
--- trunk/tests/misc/clusters.sh	2009-02-16 00:03:58 UTC (rev 2546)
+++ trunk/tests/misc/clusters.sh	2009-02-16 00:04:35 UTC (rev 2547)
@@ -1,9 +1,14 @@
 #!/bin/bash
 
-# this will turn on clustering. not sure what the best programmatic way
+# this will turn on clustering.
+
+# TODO need to check that clustering actually gets used.
+#   unsure what the best programmatic way
 # to determine that clustering actually happen is. perhaps look at
 # the cluster log file and see that there is only one of them or
 # run some log analysis to determine the clustering pattern.
+#   or, could prepend a touch onto the start of seq.sh for the
+# duration of this test, and then check that touch later
 
 export CF=swift.properties.no-retries
 
@@ -11,9 +16,18 @@
 
 cd ../language-behaviour
 
-cat $(dirname $(which swift))/../etc/swift.properties | grep --invert-match -E '^clustering.enabled=' > $CF
+cat $(dirname $(which swift))/../etc/swift.properties | grep --invert-match -E '^clustering.enabled=' | grep --invert-match -E '^clustering.min.time=' | grep --invert-match -E '^execution.retries=' > $CF
+
 echo clustering.enabled=true >> $CF
+echo clustering.min.time=600 >> $CF
 
+# turn of retries, so that if we fail in a cluster, we don't give the
+# engine another chance to run potentially outside of.
+# (in r2541, although clustering is broken, this cluster test can pass
+# because rate limiting eventually takes us down to 1 job at once, which
+# then avoids clustering, and succeeds)
+echo execution.retries=0 >> $CF
+
 export SWIFT_TEST_PARAMS="-config $CF -tc.file ../misc/clusters.tc.data"
 
 ./run 130-fmri




More information about the Swift-commit mailing list