Silence unpickle errors in accounting daemon for Archives without meta.pck

[jonasbardino]

Use allow_missing flag to unpickle call so that accounting only warns about Archives without meta.pck file rather than reporting unpickle error and warning. Then the resulting empty meta dict will result in the warning and skip as suggested in #601.

[jonasbardino]

Looking at the code it appears it only does accounting for legacy Archives in freeze_home/Archive-XYZ123, btw. That should be extended to also cover modern ones in freeze_home/USER_ID/Archive-XYZ123 if so.

[Martin-Rehr]

Looking at the code it appears it only does accounting for legacy Archives in freeze_home/Archive-XYZ123, btw. That should be extended to also cover modern ones in freeze_home/USER_ID/Archive-XYZ123 if so.

Modern archives are accounted for:

migrid-sync/mig/lib/accounting.py

Line 540 in 1bafdfe

if freeze_bytes > 0:

[Martin-Rehr]

The underlying faulty legacy archive metadata should be fixed rather than ignored

[jonasbardino]

The underlying faulty legacy archive metadata should be fixed rather than ignored

That may be the case but I still would say it's not an accounting error, so the absence of meta should at most be logged as a single warning and not repeatedly nag us about it. For recent archives you may also just stumble upon lacking meta because it's currently being created.

We have a TODO about checking archives in janitor to catch and fix the case of ones stuck in UPDATING, and checking validity of all archives would fit well there. Then it could either try to fix any broken ones or occasionally notify admins with proper back-off like we do for failing requests in #600.

[jonasbardino]

Looking at the code it appears it only does accounting for legacy Archives in freeze_home/Archive-XYZ123, btw. That should be extended to also cover modern ones in freeze_home/USER_ID/Archive-XYZ123 if so.

Modern archives are accounted for:

migrid-sync/mig/lib/accounting.py

Line 540 in 1bafdfe

if freeze_bytes > 0:

Alright, I went looking for somewhere traversing freeze_home but didn't find anything in the code. Now looking at my Account page they do appear to be counted, though.
Are the quota_home/*.pck files it parses into freeze_quota_files and iterates over filled by some external FS (Lustre) entity?

[Martin-Rehr]

The underlying faulty legacy archive metadata should be fixed rather than ignored

That may be the case but I still would say it's not an accounting error, so the absence of meta should at most be logged as a single warning and not repeatedly nag us about it. For recent archives you may also just stumble upon lacking meta because it's currently being created.

We have a TODO about checking archives in janitor to catch and fix the case of ones stuck in UPDATING, and checking validity of all archives would fit well there. Then it could either try to fix any broken ones or occasionally notify admins with proper back-off like we do for failing requests in #600.

It's (also) an accounting error since the size of the freeze archive is not accounted fore, but fine with me to leave it to the janitor to keep the freeze archives sane.

[Martin-Rehr]

Looking at the code it appears it only does accounting for legacy Archives in freeze_home/Archive-XYZ123, btw. That should be extended to also cover modern ones in freeze_home/USER_ID/Archive-XYZ123 if so.

Modern archives are accounted for:

migrid-sync/mig/lib/accounting.py

Line 540 in 1bafdfe

if freeze_bytes > 0:

Alright, I went looking for somewhere traversing freeze_home but didn't find anything in the code. Now looking at my Account page they do appear to be counted, though. Are the quota_home/*.pck files it parses into freeze_quota_files and iterates over filled by some external FS (Lustre) entity?

The design is that the quota information in quota_home is populated by the grid_quota daemon.

NOTE: For legacy FS's that we don't wan't to integrate in the grid_quota daemon (such as gluster) it's populated from an external source.

[Martin-Rehr]

Approved with the condition that the janitor will take care of fixing / screaming about broken archives.

[jonasbardino]

Looking at the code it appears it only does accounting for legacy Archives in freeze_home/Archive-XYZ123, btw. That should be extended to also cover modern ones in freeze_home/USER_ID/Archive-XYZ123 if so.

Modern archives are accounted for:

migrid-sync/mig/lib/accounting.py

Line 540 in 1bafdfe

if freeze_bytes > 0:

Alright, I went looking for somewhere traversing freeze_home but didn't find anything in the code. Now looking at my Account page they do appear to be counted, though. Are the quota_home/*.pck files it parses into freeze_quota_files and iterates over filled by some external FS (Lustre) entity?

The design is that the quota information in quota_home is populated by the grid_quota daemon.

NOTE: For legacy FS's that we don't wan't to integrate in the grid_quota daemon (such as gluster) it's populated from an external source.

Could you please point me to where grid_quota handles anything regarding archives in the non-gluster case? Because I still can't find anything or even mention of freeze_home apart from that legacy meta.pck load when browsing or greping the code in sbin/grid_{quota,accounting}.py and mig/lib/*.py.

[jonasbardino]

Approved with the condition that the janitor will take care of fixing / screaming about broken archives.

That task is now registered in #633.

[Martin-Rehr]

Looking at the code it appears it only does accounting for legacy Archives in freeze_home/Archive-XYZ123, btw. That should be extended to also cover modern ones in freeze_home/USER_ID/Archive-XYZ123 if so.

Modern archives are accounted for:

migrid-sync/mig/lib/accounting.py

Line 540 in 1bafdfe

if freeze_bytes > 0:

Alright, I went looking for somewhere traversing freeze_home but didn't find anything in the code. Now looking at my Account page they do appear to be counted, though. Are the quota_home/*.pck files it parses into freeze_quota_files and iterates over filled by some external FS (Lustre) entity?

The design is that the quota information in quota_home is populated by the grid_quota daemon.
NOTE: For legacy FS's that we don't wan't to integrate in the grid_quota daemon (such as gluster) it's populated from an external source.

Could you please point me to where grid_quota handles anything regarding archives in the non-gluster case? Because I still can't find anything or even mention of freeze_home apart from that legacy meta.pck load when browsing or greping the code in sbin/grid_{quota,accounting}.py and mig/lib/*.py.

Good catch, it doesn't seem like the freeze (archives) made it into the grid_quota daemon (yet) for lustre:

migrid-sync/mig/lib/lustrequota.py

Line 606 in 2491559

# Update quota

[jonasbardino]

Looking at the code it appears it only does accounting for legacy Archives in freeze_home/Archive-XYZ123, btw. That should be extended to also cover modern ones in freeze_home/USER_ID/Archive-XYZ123 if so.

Modern archives are accounted for:

migrid-sync/mig/lib/accounting.py

Line 540 in 1bafdfe

if freeze_bytes > 0:

Alright, I went looking for somewhere traversing freeze_home but didn't find anything in the code. Now looking at my Account page they do appear to be counted, though. Are the quota_home/*.pck files it parses into freeze_quota_files and iterates over filled by some external FS (Lustre) entity?

The design is that the quota information in quota_home is populated by the grid_quota daemon.
NOTE: For legacy FS's that we don't wan't to integrate in the grid_quota daemon (such as gluster) it's populated from an external source.

Could you please point me to where grid_quota handles anything regarding archives in the non-gluster case? Because I still can't find anything or even mention of freeze_home apart from that legacy meta.pck load when browsing or greping the code in sbin/grid_{quota,accounting}.py and mig/lib/*.py.

Good catch, it doesn't seem like the freeze (archives) made it into the grid_quota daemon (yet) for lustre:

migrid-sync/mig/lib/lustrequota.py

Line 606 in 2491559

# Update quota

Alright, not just me going crazy then :-)

Not urgent for us I suppose but perhaps an idea to register an issue to track it.

[Martin-Rehr]

Looking at the code it appears it only does accounting for legacy Archives in freeze_home/Archive-XYZ123, btw. That should be extended to also cover modern ones in freeze_home/USER_ID/Archive-XYZ123 if so.

Modern archives are accounted for:

migrid-sync/mig/lib/accounting.py

Line 540 in 1bafdfe

if freeze_bytes > 0:

Alright, I went looking for somewhere traversing freeze_home but didn't find anything in the code. Now looking at my Account page they do appear to be counted, though. Are the quota_home/*.pck files it parses into freeze_quota_files and iterates over filled by some external FS (Lustre) entity?

The design is that the quota information in quota_home is populated by the grid_quota daemon.
NOTE: For legacy FS's that we don't wan't to integrate in the grid_quota daemon (such as gluster) it's populated from an external source.

Could you please point me to where grid_quota handles anything regarding archives in the non-gluster case? Because I still can't find anything or even mention of freeze_home apart from that legacy meta.pck load when browsing or greping the code in sbin/grid_{quota,accounting}.py and mig/lib/*.py.

Good catch, it doesn't seem like the freeze (archives) made it into the grid_quota daemon (yet) for lustre:

migrid-sync/mig/lib/lustrequota.py

Line 606 in 2491559

# Update quota

Alright, not just me going crazy then :-)

Not urgent for us I suppose but perhaps an idea to register an issue to track it.

Done with issue ##635