wiki:developers-meeting-20080827
Last modified 13 years ago Last modified on 08/27/08 17:03:14

dCache developers phone conference Aug 27, 2008

Participants : Ted, Timur, Gene , Vladimir, Alex, Tigran, Patrick

Development

  • Status of patches.
  • Gerd would like to have 2 patches reviewed urgently
    • 3344 Fixes race condition in P2PClient DONE
    • 3352 Fix deadlock in pool DONE
  • CCRC fixes : Pin lifetime is after the file is staged and read pool is selected DONE
  • Patch on remove files in repository.
  • Alternative Pool Selection mechanism.
  • Developers should become familiar with Gerds new migration module in pool.

Tickets

  • IN2P3 : gPlazma question by Jonathan (Ted) ? Done.
  • Flavia : starting pin timer : on bring-online or when file touches the disk. Done.
  • Release pin by VO. On the way. Need changes at gPlazma and SRM. Planned for Sep.
  • 3391 By Tanya : SrmSpaceManager doesn't clean up failed transfer from srmspacefile table
  • Grid ftp 'ls -l' output format ? Gerd would like to do this. Are there any objections ? Ask Gerd to check standards. Ask Gerd if he wants to add the dcls command as well.

Operational

  • NDGF is upgrading to chimera : September 15th to 19th
  • Jon : Some Doors just hang. No more information yet. Timur will do full investigation. No answer yet.
  • Proposal
    • Add source of messages when deleting files on pools. Timur is working on this one. POOLS didn't remove files. Was human or script.
    • Report all commands on the Admin cell. Should be already implemented.
  • Ted OSG Operation : Chris Green (Fermilab is hosting the central server for collecting probe information) : Indexing the billing database. Brian provided the 'sql' query. Indexing indeed required. Ted will send procedure to sys admin on how to do the indexing of the 'time'.
  • Doris : 2 issues (independent)
    • Box like peak in the pnfs manager queue. Doesn't influence production. All requests are 'getMetaData'. Silke provided the PnfsManager queue dump. The getMetadata messages are coming from the SRM cell. So very likely an 'ls' storm by FTS.
    • Sawtooth like peak in the pnfs manager queue. Slows down the system. Result : Many SyncWait on SRM node. Postgres log shows that the postgres request time is increasing from 250ms to several seconds. So either postgres is doing something weird or the underlying file system is periodically very slow. From SAR Paul could create some plots which confirms that massive reading is going on before the sawtooth peak begins. No information yet from Silke what that could be.
  • BNL reported pnfs problems. Turned out that dbserver '0022' has died. Nobody at BNL has been able to figure this out. pnfs and cache core had been restarted randomly when the problem appeared.
  • Authorization/Authentication?. Slow start feedback on Ted's proposal.
    • Patrick went to CMS software meeting in Aachen to collect requirements.
  • Future :
    • one UID multiple group ID's.
    • One DN/Role/Group(FQAN) may have multiple gid's.
    • User Record : one uid, list of list of gid's, one user root directory.
    • Still dicussion on the gPazma migration.
    • Auto-generation of UID.
  • Gerd Question : Gerd is going to the atlas jamboree. (still)
  • Gerd Question : New Queue for PRODUCTION PATCH. Not a new queue but a tag. Patrick will provide more info later.

General

  • We are still preparing a 5 hours dCache tutorial for the gridKa school. We are not yet panicing, though we are not far from doing so. The European school for grid newcomers.

Last Modified by patrick @ Sat Feb 27 16:59:14 2021