Concurrency issues in o.a.i.hive.MetastoreLock

[snazy]

Apache Iceberg version

None

Query engine

None

Please describe the bug 🐞

org.apache.iceberg.hive.MetastoreLock#acquireJvmLock can race with releaseJvmLock and produce a NullPointerException. Additionally, the implementation has concurrency issues.

The call to Thread.interrupted() in the InterruptedException catch blocks is meaningless.

[pvary]

@snazy: I don't think there is an issue here. New MetastoreLock is created for every commit operation, and they are not reused or shared.

[snazy]

@pvary hm - but why's there a "lock" and "unlock" operation, if it's not shared?
The comment in lock() says it is shared though: avoid concurrent commit attempts to the same table.
If it is shared, it has concurrency issues:

• race on jvmLock field between releaseJvmLock() and acquireJvmLock
• visibility issue of the jvmLock field

[pvary]

The cache is shared, but the lock attribute of the MetastoreLock is not shared.