Hello Michal and Andy Thanks for the explanation. I think everything is fine and the issue is due to the way I use TPC. On the destination site the xrdcp that does the actual file transfer writes the file first to a temporary name and renames the file only after the transfer finishes successfully. My understanding is that due to this the sync request will timeout for long lasting transfers. I will change our tools to have the client do the temporary-name/rename steps. Would it be possible to update the xrdcp man page and explain CPTPCTimeout a little bit more. Cheers, Wilko On Thu, 14 Dec 2017, Michal Kamil Simon wrote: > Hi guys, > > In order to initiate the TPC the client needs to issue a sync request against the destination file, > if the sync request times out (request timeout) the whole TPC operation fails. This simply means > the client cannot reach the destination within the specified request timeout. The TPC timeout > doesn't apply as the TPC hasn't been initiated yet. > > I think this behaviour is rather correct. > > Cheers, > Michal > ________________________________________ > From: [log in to unmask] [[log in to unmask]] on behalf of Andrew Hanushevsky [[log in to unmask]] > Sent: 14 December 2017 09:59 > To: Wei Yang > Cc: Kroeger, Wilko; xrootd-l; Michal Simon > Subject: Re: increasing allowed time for xrdcp third party copy > > No, Wilko was actually hitting the request timeout. It really should be > better handled and I'll talk to Michal about that. The TPC timeout > actually refers to how long the TPC rendezvous can take. After that, the > request timeout applies. So, the documentation is misleading. > > Andy > > On Thu, 14 Dec 2017, Yang, Wei wrote: > >> I wonder if TPC should send out heartbeat signal, like what GridFTP does. >> >> -- >> Wei Yang | [log in to unmask] | 650-926-3338(O) >> >> >> >> >> >> >> >> -----Original Message----- >> From: <[log in to unmask]> on behalf of Wilko Kroeger <[log in to unmask]> >> Date: Thursday, December 14, 2017 at 2:39 AM >> To: xrootd-l <[log in to unmask]> >> Subject: increasing allowed time for xrdcp third party copy >> >>> Hello >>> >>> We are using the third party copy option (--tpc only) for xrdcp to >>> transfer files between two xrootd clusters and noticed that transfers >>> failed if they took longer than 30 min (1800s). >>> >>> We could fix this issue by increasing the request timeout: >>> XRD_REQUESTTIMEOUT=7200 >>> (default 1800) but according to the documentation it seems like >>> XRD_CPTPCTIMEOUT should be used in this case: >>> XRD_CPTPCTIMEOUT (-DICPTPCTimeout) >>> Maximum time allowed for a third-party copy operation to finish. >>> However increasing or decreasing it had no effect on the transfers. >>> Is this expected or is there an issue with xrdcp how it handles these two >>> timeouts? >>> >>> >>> Here is an example of a failed transfer (tested with v4.6.1 and 4.8.0-rc2): >>> >>> 2017-12 13 15:01:51 -0800 xrdcp --debug 1 --nopbar --tpc only root://..... root://..... >>> >>> [2017-12-13 15:31:51.940965 -0800][Error ][XRootD ] [trg-xrootd-srv] Unable to get the response to request kXR_sync (handle: 0x00000000) >>> [2017-12-13 15:31:51.941069 -0800][Error ][File ] ...... Fatal file state error. Message kXR_sync (handle: 0x00000000) returned >>> with [ERROR] Operation expired >>> [2017-12-13 15:31:52.376810 -0800][Error ][Utility ] Third party copy from .... to ... failed: [ERROR] Operation expired >>> [2017-12-13 15:31:52.379876 -0800][Error ][XRootDTransport ] Message 0xb0000d50, stream [1, 0] is a response that we're no longer interested in >>> (timed out) >>> Run: [ERROR] Operation expired >>> >>> >>> Cheers, >>> Wilko >>> >>> ######################################################################## >>> Use REPLY-ALL to reply to list >>> >>> To unsubscribe from the XROOTD-L list, click the following link: >>> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1 >> >> ######################################################################## >> Use REPLY-ALL to reply to list >> >> To unsubscribe from the XROOTD-L list, click the following link: >> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1 >> > > ######################################################################## > Use REPLY-ALL to reply to list > > To unsubscribe from the XROOTD-L list, click the following link: > https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1 ######################################################################## Use REPLY-ALL to reply to list To unsubscribe from the XROOTD-L list, click the following link: https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1