I am seeing what looks like inconsistent reporting by stat (2).
Below is the annotated trace that shows a successful stat of the
file "part-m-00001", and then later a failed stat of the same file.
This trace gathered with strace. The scenario can be fully reproduced, but so far
not with any exemplary, concise code; this trace comes deep from within Java.
Setup: two files exist under the Ceph mount point (/ceph):
/ceph/ts5-in/part-m-00000
/ceph/ts5-in/part-m-00001
Link to full trace: http://www.cs.ucsc.edu/~jayhawk/log2-full.txt.tar.gz
-------
(1) In the first part of the trace a thread successfully stats both files,
and performs an ioctl on both files:
--> 10941 stat("/ceph/ts5-in/part-m-00000", {st_mode=S_IFREG|0644, st_size=2500000, ...}) = 0
--> 10941 stat("/ceph/ts5-in/part-m-00001", {st_mode=S_IFREG|0644, st_size=2500000, ...}) = 0
10941 open("/ceph/ts5-in/part-m-00000", O_RDONLY) = 55
10941 ioctl(55, 0x80289701, 0x7fd7a7efd9e0) = 0
10941 ioctl(55, 0xc0f89703, 0x7fd7a7efda10) = 0
10941 close(55) = 0
10941 open("/ceph/ts5-in/part-m-00001", O_RDONLY) = 55
10941 ioctl(55, 0x80289701, 0x7fd7a7efd9e0) = 0
10941 ioctl(55, 0xc0f89703, 0x7fd7a7efda10) = 0
10941 close(55) = 0
...
(2) Later in the trace, a thread (10956) stats "part-m-00000" with success,
while the thread (10957) stats "part-m-00001" and fails with ENOENT.
10956 stat("/ceph/ts5-in/part-m-00000", <unfinished ...>
10950 <... futex resumed> ) = 0
10957 write(2, "\n", 1) = 1
10950 gettimeofday({1294621240, 685789}, NULL) = 0
--> 10957 stat("/ceph/ts5-in/part-m-00001", <unfinished ...>
10950 gettimeofday( <unfinished ...>
--> 10957 <... stat resumed> 0x7fd75db461f0) = -1 ENOENT (No such file or directory)
10956 <... stat resumed> {st_mode=S_IFREG|0644, st_size=2500000, ...}) = 0
Thanks,
Noah--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
