yahoo-eng-team team mailing list archive
-
yahoo-eng-team team
-
Mailing list archive
-
Message #12162
[Bug 1297451] [NEW] Image uploads to the filesystem driver are not fully atomic
Public bug reported:
When uploading to the filesystem store, the image is uploaded,
checksummed, and added to the database, with appropriate rollback
operations in place in case of failures. This is all good, but there
are various corner-cases where partially-uploaded files can be left
lying around in the filesystem_store_datadir.
Mostly this kind of thing interferes with operational processes we have
that are external to Glance and not Glance itself. e.g., we have a
monitoring process to reconcile the list/checksums of glance images
against what's on-disk, and this occasionally goes funny on us.
Additionally, we have a side-band backup/replication process that ends
up occasionally transferring partial files for no reason.
Using a temporary upload space and moving the file into place on the
filesystem store would cause a lot of our co-processes to work more
cleanly, as well as reducing the risk of the _delete_partial job
deleting the wrong thing should it go sideways.
** Affects: glance
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to Glance.
https://bugs.launchpad.net/bugs/1297451
Title:
Image uploads to the filesystem driver are not fully atomic
Status in OpenStack Image Registry and Delivery Service (Glance):
New
Bug description:
When uploading to the filesystem store, the image is uploaded,
checksummed, and added to the database, with appropriate rollback
operations in place in case of failures. This is all good, but there
are various corner-cases where partially-uploaded files can be left
lying around in the filesystem_store_datadir.
Mostly this kind of thing interferes with operational processes we
have that are external to Glance and not Glance itself. e.g., we have
a monitoring process to reconcile the list/checksums of glance images
against what's on-disk, and this occasionally goes funny on us.
Additionally, we have a side-band backup/replication process that ends
up occasionally transferring partial files for no reason.
Using a temporary upload space and moving the file into place on the
filesystem store would cause a lot of our co-processes to work more
cleanly, as well as reducing the risk of the _delete_partial job
deleting the wrong thing should it go sideways.
To manage notifications about this bug go to:
https://bugs.launchpad.net/glance/+bug/1297451/+subscriptions
Follow ups
References