Self healing of missing accounts/containers

[David Hadas]

Self healing is not implemented when an account is "lost".
Same for containers.

If an account is missing from the system but we have a container indicating that the account exists - the container updater would try to update the account with PUT /missingaccount/existingcontainer which fails on 404
Same for containers.

Suggested behavior for accounts:
1. Restore the account
2. Mark the account as restored while marking its ACLs to be restrictive - no one can list/read/write containers in this account until the next PUT or POST on the account setting proper ACLs.

Suggested behavior for containers:
1. Restore the container
2. Mark the container as restored while marking its ACLs to be restrictive - no one can list/read/write objects in this container until the next PUT or POST on the container setting proper ACLs.

[Samuel Merritt (torgomatic)]

Under what circumstances are there orphaned containers? I'd be more interested in preventing that from happening than in making up some resurrected account of dubious accuracy.

[Matthew Oliver (matt-0)]

I agree with Sam, what is the definition of "lost" here, to be lost _all_ replicated versions of it have to be misplaced before a replicator cycle has run... if there are good failure domains etc, then this is something swift tries really hard to avoid... its kind of the point of swift.

So in a way, replication and the replicator does in fact make swift self healing.

Unless lost can be defined, I'm thinking this bug should change to won't fix or invalid.

[Matthew Oliver (matt-0)]

Marking as incomplete, until "lost" is defined.

[Launchpad Janitor (janitor)]

[Expired for OpenStack Object Storage (swift) because there has been no activity for 60 days.]