Lease renewal behavior

Lionel Bouton lionel-subscription at
Tue Apr 22 14:27:22 UTC 2014


I was wondering how the repair and lease renewal system behaves in the
following situation.

The storage network is mostly used for long-term archives, garbage
collecting is active on storage nodes and lease renewal/repair is used
to maintain an healthy availability. All storage nodes are maintained by
a single administrator.
1/ files are stored on the storage network,
2/ some of the nodes go offline but don't lose their data store: some
shares are now unavailable and availability doesn't reach the "healthy"
threshold for some files,
3/ a repair happens while the nodes are offline and shares of offline
servers are regenerated from available shares and put on available
storage nodes,
4/ the offline nodes become available again: at this point for each
repaired share the whole LAFS storage network now have a duplicate,
5/ a lease renewal happen.

My question is what happens exactly when the lease renewal is processed:
- are every duplicate share renewed or is only one of them renewed ?
- if only one is renewed, is the algorithm choosing the storage node on
which renewal is done for the duplicated share deterministic ?
And finally if only one share is renewed but the algorithm choosing it
isn't deterministic is there anything the administrator of the whole
network can do to make it deterministic ?

The underlying question is related to storage efficiency in this
context: is space "wasted" by duplicate shares that are always renewed
and won't be garbage collected until the file they belong to is deleted ?

Best regards,

Lionel Bouton

More information about the tahoe-dev mailing list