Dear dCache team,
Could you please implement the prepare XRootD command? In ALICE we use this protocol exclusively and it would streamline the upcoming recall campaign in LS3.
Currently the transfers work by blindly trying to open the file for reading. And waiting for either to work or timeout. This does trigger an internal prepare and retrying these transfers they eventually work. The procedure is inefficient because we cannot queue a certain number of files (function of the buffer size) and thus will trash the buffer before they are retried and can be transferred.
If we can rely on the disk buffer capacity (subject to this other ticket : #8105) and prepare works then we can queue just as many files as the buffer can take.
Since the operation is anyway triggered internally and there is an implementation of prepare, I hope the two pieces of code can be connected and we also get this working.
Thanks a lot for your support,
.costin
Dear dCache team,
Could you please implement the
prepareXRootD command? In ALICE we use this protocol exclusively and it would streamline the upcoming recall campaign in LS3.Currently the transfers work by blindly trying to open the file for reading. And waiting for either to work or timeout. This does trigger an internal
prepareand retrying these transfers they eventually work. The procedure is inefficient because we cannot queue a certain number of files (function of the buffer size) and thus will trash the buffer before they are retried and can be transferred.If we can rely on the disk buffer capacity (subject to this other ticket : #8105) and
prepareworks then we can queue just as many files as the buffer can take.Since the operation is anyway triggered internally and there is an implementation of
prepare, I hope the two pieces of code can be connected and we also get this working.Thanks a lot for your support,
.costin