wiki:developers-meeting-20071004
Last modified 13 years ago Last modified on 10/05/07 13:22:08

Developers Phone conference Oct 4

Participants : Timur, Gerd, Tigran, Martin, Gene, Dmitry, Alex (later), Valdimir, Patrick

Planing :

  • Upgrade of the Tier I's to dCache 1.8. No further comments except that we should try to keep the break between the fzk upgrade and the IN2P3 in order to fix issues reported by NDGF and fzk.
  • srm client tool support : Timur : We assume that clients as they currenlty are, are fine and don't need significant work done on them any more. If more work is needed, all other SRM work will have higher priority.
  • dCache 1.8 and resilient management : There are two streams of 'resilient work'.
    • The current Resilient Manager is slightly improved and will be made usable with 1.8. This RM will not be able to mix resilient and none resilient pools. Furthermore it will not be able to understand spaces. This RM will (theoretically allow Jon to upgrade the USCMS system within the WLCG schedule)
    • A new RM will be introduced, which, per design, will fit into the new dCache 1.8 Space functionality.
  • Our contribution to the GSSD meeting Tuesday at CERN : Timur and Dmitry will prepare slides which patrick will present. Timur will be present during the afternoon meeting via VRVS. The slides will be available Friday night. So that it is too late for Patrick to comment on them :-).
  • GLUE schema discussion, summarized by Greg. (Please see attachement) Timur will comment on it and hopefully will prepare slides.
  • We are going to cut a new 1.8 release on monday if Dmitry reports on BUG FIXes of 'Flavia issues'. Otherwise the current release (from today) will do.

Development :

  • Open 'flavia' issues ? There are two kind of issues. A) Those which have been fixed in the currenlt release and b) those for which Dmitry has been sending a request for clarification to Flavia. All contravorsy issues will be reported on by Timur/Dmitrys? slides for the GSSD.
  • Open important issues ?
    • Files remain on pool after being removed by srm-advisory-delete. Seems to be fixed after Timur removed the 'flags' message. Tanya will repeat the tests and will report on the results.
    • Disk space management on pools. After the 'enable' message has been removed from the RM code, the situation improved, though the basic effect has been observed. Timur will gather more information and will check with the GFRP mover.
    • Out of memory for grid ftp doors. One of the reasons is that if a client siliently dies, the door stucks in a 'join' command because the Adapters are still waiting for the final finish of the transfer. This will be adressed by merging the NDGF SocketAdapater which
      • if a transfer 'ok' is reported by the pool, the door waits for some time and if not all threads of the adapter are gone, it kills the remaining ones.
      • if a transfer 'bad' is reported by the pool the adapter is killed.
      • the new adapter, other than the old one, understands and honours java interrupts.
  • Alex will talk about the Resilient Manager usage of the pool functionality and we would like to hear your suggestions on how to put pool in "drain off" state without using the disable functionality. Setting rdOnly is suggested. Alex will try this.
  • Changes between the PoolManager configuration and the SpaceManager configuration (boolean flags). Timur checked in the required and agreed part (and slightly more). Tigran will have his changes done mid of next week which is one week later than agreed for which I sincerely appologize.

CDF and CMS reports

  • Report from the upgrade of the CDF system (Wednesday, Oct 3). Some problems have been reported due to the fact that some pools didn't have the data portion mounted. A real issues is obviously that sometimes the p2p transfer 'nearly' suceeds but at some point (at the end) failes. This leaves zombies behind and lets dCache create more and more copies. Alex will add the output of the p2p client to the support ticket.
  • Timur would like to provide an update on the status of the CMS lfs precious pools after the usage of the enable/disable functions was turned off.


Last Modified by Patrick @ Sun Mar 7 00:23:26 2021

Attachments