Contents:

Master Recovery

Goal: No application can (overtly) distinguish that a failure has occurred using the specified client interface while less than f uncorrected failures are ongoing - with high probability.
Goal: Client can expect latency less than l - with high probability; they can further expect latency bounded by l' while the system has less than f ongoing, uncorrected failures.

...

Segment List - which segments are live for a given master
Restoration: Have to send a query to all outstanding backups. This is probably okay because we can piggyback this with the fetch of the metadata summaries. Additionally, since we are doing a broadcast to find segment backup locations we can do this with the same query.

SafeVersion (Master vector clock) - the version stamp number for the next operation object creation on this server
Restoration: After re-reading segment metadata from backups fetch the most recent value written to the most recent segment and set the vector clock one higher Reading special object 'objectSafeVersion' appended at the beginning of log segment to restore safeVersion when the segment was created. Then at the log replay, safeVersion is updated to maintain safeVersion is bigger than the version numbers of any replayed tomb stones.

Id -> Ptr Hashtable - the location in RAM for each object id
Restoration: After re-reading segment metadata from backup scan all objects, populating the hashtable.

...