Add undelete method to BlobStore

[justinlin-linkedin]

Adding undelete method to blobstore and other Store implementation.
This PR will have no change w.r.t AmbryRequests.

[codecov-io]

Codecov Report

Merging #1373 into master will increase coverage by 0.05%.
The diff coverage is 70%.

@@             Coverage Diff              @@
##             master    #1373      +/-   ##
============================================
+ Coverage     72.01%   72.06%   +0.05%     
- Complexity     6779     6791      +12     
============================================
  Files           488      488              
  Lines         38445    38504      +59     
  Branches       4890     4898       +8     
============================================
+ Hits          27685    27747      +62     
  Misses         9421     9421              
+ Partials       1339     1336       -3

Impacted Files	Coverage Δ	Complexity Δ
...in/java/com.github.ambry.cloud/CloudBlobStore.java	81.05% <0%> (-0.29%)	81 <0> (ø)
...main/java/com.github.ambry.store/StoreMetrics.java	97.26% <100%> (+0.05%)	10 <0> (ø)	⬇️
...n/java/com.github.ambry.store/PersistentIndex.java	89.15% <100%> (+0.22%)	244 <1> (+2)	⬆️
...ithub.ambry.commons/LoggingNotificationSystem.java	45.16% <12.5%> (-13.18%)	9 <1> (ø)
...rc/main/java/com.github.ambry.store/BlobStore.java	89.06% <78.72%> (-0.65%)	107 <3> (+5)
...java/com.github.ambry.store/CompactionManager.java	89.88% <0%> (-2.39%)	25% <0%> (ø)
...ain/java/com.github.ambry.router/PutOperation.java	91.73% <0%> (+0.49%)	115% <0%> (+1%)	⬆️
...c/main/java/com.github.ambry.network/Selector.java	79% <0%> (+0.78%)	78% <0%> (+1%)	⬆️
... and 6 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update adc32f9...60e973c. Read the comment docs.

[jsjtzyy]

Initial comments, will take a second look.

[jsjtzyy]

Any reason why we don't pass in MessageWriteSet? Is this because we undelete only one blob at a time?

[justinlin-linkedin]

MessgeWriteSet includes a serialized undelete record, which will be persisted on disk in the log. The problem is that this serialized record requires a lifeVersion, but we don't know the lifeVersion before we query the index. So here I pass a MessageInfo instead of MessageWriteSet. You can see in the BlobStore.undelete method, I create a MessageWriteSet in the end, with a lifeVersion.

[jsjtzyy]

Note that, other methods (put, ttlUpdate, delete) in BlobStore are also invoked in Replication. When you implement undelete replication part, the LIFE_VERSION_FROM_FRONTEND check here probably needs to change.

[justinlin-linkedin]

I actually have to change ttlupdate and delete later to make them aware of undelete in replication.

[jsjtzyy]

LGTM, one minor comment.

[zzmao]

LGTM.

One minor comment.