Summary: with big file, rsync times out out when it should not;
the sender is still responsive
AssignedTo: [hidden email] ReportedBy: [hidden email] QAContact: [hidden email]
Description of problem:
When sending big file (eg. > 500MB) With --timeout=xx option, the
server side of rsync timeouts out when it should not. The components
of rsync (generator or receiver or server) should exchange periodic
keepalive msg, so they will not timeout when the connection is
Version-Release number of selected component (if applicable):
With --timeout=10: sometimes with 500MB file; almost always with 2GB file.
Steps to Reproduce:
1. Assume timeout=10 sec. generate the source file, eg.
time dd if=/dev/zero of=$i bs=1M count=1000
If it takes less than <timeout> x 3 (eg. 30 sec), increase
the file size. Bigger file better illustrates this issue.
rsync --timeout=10 localhost:$i $j
Saw the following errors, depending on the mode of failure.
[sender] io timeout after 10 seconds -- exiting
rsync error: timeout in data send/receive (code 30) at io.c(140)
rsync: writefd_unbuffered failed to write 6 bytes to socket [generator]:
Broken pipe (32)
rsync error: timeout in data send/receive (code 30) at io.c(1530)
rsync error: received SIGUSR1 (code 19) at main.c(1288) [receiver=3.0.7]
The sync should complete without error; certainly no timeout.
1. The issue is more prounced with two hosts of different speed,
eg. h1 (dest) is much slower than h2 (src) (eg. h1 could be an
slower machine (especially I/O bound) with LUKS; this could be
simulated by strace on h1:receiver).
h2: dd if=/dev/zero of=$i bs=1M count=1000
h1: rsync --timeout 10 h2:$i $j
2. From strace, the scenario with error msg in #2:
- generator <g> read <dest> and sends checksum to
sender <s> (--server in this case).
- <g> spawns receiver <r>
- <r> opens <dest> and <tmp> file, start copying from <dest> to <tmp>
- <s> reads <src> file and sends msgs to <r>
- when <s> finishes, it starts to wait, for (timeout/2) sec.
- Assume <r> is slower, <s> timeouts out after <timeout> sec.
- <g> probably needs to send keepalive msg to <s>.
3. strace for #1 is similar to #2:
- after <s> finishes, <r> takes more than 10 sec to finish
copying <dest> to <tmp> and then rename <tmp> <dest>,
so <s> times out.
4. The issue is somewhat reproducible on one single host. But
with strace on both <s> and <r>, it becomes harder to
reproduce, probably because strace slows both <s> and <r> down.
I need to use a much bigger file (2.5GB), and strace <r> from
beginning, but strace <s> only torwards the end, to show this
error with strace.
5. There are several existing bugs that may be related to this
unexpected timeout issue.
Random high loads during syncs (server side) / client stream errors
rsync: connection unexpectedly closed (2761332 bytes received so far)
rsync error: error in rsync protocol data stream (code 12) at io.c(365)
rsync: writefd_unbuffered failed to write 4092 bytes [sender]:
Broken pipe (32)
io timeout after 30 seconds -- exiting
rsync error: timeout in data send/receive (code 30) at io.c(239)
improve keep-alive code to handle long-running directory scans
write last transfer status when timeout or other error happens
rsync: writefd_unbuffered failed to write 4 bytes [sender]:
Broken pipe (32)
rsync: connection unexpectedly closed (99113 bytes received so
rsync error: unexplained error (code 255) at io.c(600) [sender=3.0.5]
timeout reached while sending checksums for very large files
--- Comment #2 from Loïc Gomez <[hidden email]> 2012-10-23 11:26:55 UTC ---
I ran into a similar issue recently while transferring large files (>40GB).
After a few tests, it seems - in my case at least - to be related to the
delta-xfer algorithm. The bug does not happen anymore with the -W option.
I don't know if this will resolve your issue, but you can also try looking into
these options : --no-checksum --no-compress --blocking-io. These were not the
source of my problems, but the functions they're related to might rise a
I hope it helps, anyways, good luck solving your issue.
--- Comment #4 from Dominic Raferd <[hidden email]> ---
I can confirm that this bug still exists when using compression (-z) with rsync
3.1.0 (server and dest), and that removing compression is a good workaround.