Repository navigation
Description
I'm not sure if it's bug. It's more a question. As we can see,
pulsar/pulsar-client/src/main/java/org/apache/pulsar/client/impl/HandlerState.java
Lines 53 to 56 in 6089292
| protected boolean changeToReadyState() { | |
| return (STATE_UPDATER.compareAndSet(this, State.Uninitialized, State.Ready) | |
| || STATE_UPDATER.compareAndSet(this, State.Connecting, State.Ready) | |
| || STATE_UPDATER.compareAndSet(this, State.RegisteringSchema, State.Ready)); |
HandlerState#changeToReadyState is not an atomic operation. I'm not sure there's a race case like following timeline
| Time | Event | State Before | State Now |
|---|---|---|---|
| 1 | STATE_UPDATER.compareAndSet(this, State.Uninitialized, State.Ready) |
State.Connecting |
State.Connecting |
| 2 | setState(State.Uninitialized) |
State.Connecting |
State.Uninitialized |
| 3 | STATE_UPDATER.compareAndSet(this, State.Connecting, State.Ready) |
State.Uninitialized |
State.Uninitialized |
| 4 | STATE_UPDATER.compareAndSet(this, State.RegisteringSchema, State.Ready) |
State.Uninitialized |
State.Uninitialized |
As we can see, there's a time point that the state was changed back to Uninitialized from Connecting. However, we should expect the state to be Ready because neither Uninitialized nor Connecting was a closed state.
I see references of changeToReadyState in ProducerImpl and ConsumerImpl were protected by the lock directly or indirectly, like
pulsar/pulsar-client/src/main/java/org/apache/pulsar/client/impl/ConsumerImpl.java
Lines 738 to 739 in 6089292
| synchronized (ConsumerImpl.this) { | |
| if (changeToReadyState()) { |
I'm not sure if the lock works because it requires some setState invocations are protected by the lock and I didn't check it in detail.
And in TransactionMetaStoreHandler#connectionOpened, there's no lock.
pulsar/pulsar-client/src/main/java/org/apache/pulsar/client/impl/TransactionMetaStoreHandler.java
Lines 115 to 117 in 6089292
| if (!changeToReadyState()) { | |
| cnx.channel().close(); | |
| } |
I'm not sure if the thread safety could be guaranteed. IMO, if there's no possibility that the state was changed back to Connecting or Uninitialized during changeToReadyState, it will be thread safe. Or this race condition is acceptable?
Activity
[-]Thread safe problem about HandlerState#changeToReadyState[/-][+][QUESTION] Thread safe problem about HandlerState#changeToReadyState[/+]BewareMyPower commentedon Mar 20, 2021
Is following implementation better?
codelipenghui commentedon Mar 4, 2022
The issue had no activity for 30 days, mark with Stale label.