Closed
Bug 307573
Opened 19 years ago
Closed 16 years ago
prpollml test failures
Categories
(NSPR :: NSPR, defect)
Tracking
(Not tracked)
RESOLVED
FIXED
4.7
People
(Reporter: jason.m.reid, Assigned: wtc)
Details
Attachments
(2 files, 1 obsolete file)
89.91 KB,
text/plain
|
Details | |
2.06 KB,
patch
|
wtc
:
review+
|
Details | Diff | Splinter Review |
prpollml test fails on HP-UX and Solaris. Test fails immediately with: BEGIN TEST: prpollml PR_NewTCPSocket failed END TEST: prpollml Occurs regardless of OPT/DBG or 32/64 bit.
Reporter | ||
Comment 1•19 years ago
|
||
Assignee | ||
Comment 2•19 years ago
|
||
This should be fixed by increasing the "descriptors" limit of the shell to 512 in runtests.sh. prpollml needs to create more than 256 sockets. If the default "descriptors" limit is <= 256, the test fails.
Summary: prpollml test failures → prpollml test failures
Updated•18 years ago
|
QA Contact: wtchang → nspr
Comment 3•16 years ago
|
||
Wan-teh certain OS's will not allow the user's shell to increase the per process file descriptor. One must configure the machine as root. Would it be acceptable to decrease POLL_DESC_COUNT to 225 or 192 since the value needs to be greater than STACK_POLL_DESC_COUNT in ptio.c which is currently set to 64. This would remove prpollml failure from several QA machines rather than having to configure the QA machines.
Attachment #298999 -
Flags: review?(wtc)
Comment 4•16 years ago
|
||
Comment on attachment 298999 [details] [diff] [review] decrease POLL_DESC_COUNT This patch is ok, except for its reduction in the poll descriptor count. Glen, descriptor limits less tan 256 seem ridiculously low. There's no way that any server product should ever be expected to run on a machine with such low limits. Let's get the QA machines configured with reasonable limits, rather than hobbling NSPR's tests.
Attachment #298999 -
Flags: review-
Comment 5•16 years ago
|
||
informing Christophe and Slavo that we should configure the QA machines instead.
Comment 6•16 years ago
|
||
Attachment #298999 -
Attachment is obsolete: true
Attachment #299040 -
Flags: review?(wtc)
Attachment #298999 -
Flags: review?(wtc)
Assignee | ||
Comment 7•16 years ago
|
||
Comment on attachment 299040 [details] [diff] [review] increase the file descriptor limit r=wtc. 1. Add (int) casts in front of PR_GetError() and PR_GetOSError() because these functions return PRInt32 but the %d printf format specifier expects an int. 2. These two fprintf statements can be merged into one: >+ fprintf(stderr, "Ensure the per process file descriptor limit " >+ "is greater than %d", POLL_DESC_COUNT); The C compiler will concatenate string literals for us. Note that you had an extra space at the beginning of the second string literal. 3. In runtests.sh, change "Borne" to "Bourne". 4. You can use 512 instead of 1024. A descriptor limit of 512 is sufficient to run the prpollml test. 5. Could you CVS remove runtests.ksh? Thanks!
Attachment #299040 -
Flags: review?(wtc) → review+
Comment 8•16 years ago
|
||
/cvsroot/mozilla/nsprpub/pr/tests/prpollml.c,v <-- prpollml.c new revision: 3.7; previous revision: 3.6 done Checking in runtests.sh; /cvsroot/mozilla/nsprpub/pr/tests/runtests.sh,v <-- runtests.sh new revision: 1.6; previous revision: 1.5 done
Status: NEW → RESOLVED
Closed: 16 years ago
Resolution: --- → FIXED
Assignee | ||
Updated•16 years ago
|
Target Milestone: --- → 4.7
You need to log in
before you can comment on or make changes to this bug.
Description
•