[part of a series of meetings]

Participants

Patrick, Paul, Karsten, Tanja; Gerd; Dmitry, Gene

Agenda

[see box on the right-hand side]

Postcards

Up to two minutes (uninterrupted) per person where they can answer two questions:

• What I did last week (since the last meeting),
• What I plan to do in the next week.

No questions until we get through everyone :)

Tanja: ticket, patches, review board, poo2pool via HTTP. Have been looking at the pool2pool state-machine; problem with GGUS tickets (currently we only get reminders, which is bad).

Gerd: Spent a bit of time helping Shawn McKey?: failure freq. dropped when changing JVM but still there; Zurich talking to drop-box competitor company, nothing fruitful at this time. Spent time working on pool-manager links expressions .

Dmitry: worked a little bit on project management (tactical plan for the next year), continuing browsing/fixing SRM code. Adding Data Nucleus for persistence mechanism for requests .. reflecting object heirachy in database.

Gene: helping with the tactical plans.

Patrick: paperwork ..

Paul:

Karsten: took over EMI testbed over while Christian was away .. lots of fun.

Plans for patch-releases

Should we make a new patch release?

Please don't delay release on test.

Dmitry is happy to test 1.9.12 with PNFS.

Trunk activity

Progress with new features...

Pools

Pool in an overload situation: adding a new mover decreases the throughput. How do we recognise this?

Dima's proposed soln is to use p2p to copy files out of the slow pool

Issues from yesterday's Tier-1 meeting

GridKa : pool read-only mode

Patch to allow a pool to start in pure read-only mode: on a read-only filesystem.

Issues from EMI

Nagios probes: a generic SRM probe and a dCache-specific one.

S2 tests / SRM

Action on no put Done?

DPM deletes data; Castor and dCache "auto-commit" the data.

What does the spec. say?

A way to see if VOs are relying

Implement the file shouldn't be readable until srmPutDone. If we start enforcing this then this would check whether people are "cheating" without deleting their data.

Do we log TURLs timing out? Don't think so.

Right now the SRM overwrite option: if enabled, doesn't

We have to enable overwrite in the doors: would be better if the SRM did the delete. Easy to fix.

FTS does the delete before upload.

ATLAS mgmnt

Asking for advise.

Does it make sense to delete directories if they don't contain data?

Another possibility is to use the msg-system to report that a directory is empty.

Could modify the syncat format to include directories?

They're doing using srmLs to scan an entire space to find out if a directory is empty.

Auto-generated code

Just fix the client to validate input.

Outstanding RT Tickets

[This is an auto-generated item. Don't add items here directly]

 RT 6463: feature request: statistics for pinned data in pools

 RT 6493: Data lost when migration from pnfs to chimera

What is the event to remove the has-been-removed entry in the trash table? say ~1/2 year? We don't need to be too aggressive as: 1. only the PNFS-ID need be stored, 2. the table will only lightly used, so won't use the available cache.

Fix the documentation to make it clear.

Adjust "pnfs register" command in a pool so, by default, it doesn't delete files if a namespace entry isn't present.

Add support for a "--delete-orphans" option to provide original behaviour; maybe also a "--dry-run" option to report which files would be deleted.

On pool-startup, files may be deleted. An entry marked "broken" will ... Try to start a pool will try to register the file: this may delete the file.

If someone deletes the metadata and starts the pool: any missing entries will be deleted.

Review of RB requests

DTNM

Proposed: same time, next week.