[
https://issues.apache.org/jira/browse/IGNITE-18091?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17634256#comment-17634256
]
Denis Chudov edited comment on IGNITE-18091 at 11/15/22 9:28 AM:
-----------------------------------------------------------------
*Correct* means that the behavior matches the expectations.
||N||Scenario||Ignite 3||MVP||Expected||
|1|*testWaitDie0*
var tx1 = beginTx();
var tx2 = beginTx();
var key1 = key("test");
xlock(tx2, key1).join();
CompletableFuture<?> xlockFutTx1 = xlock(tx1, key1);
assertFalse(xlockFutTx1.isDone());
commitTx(tx2);
xlockFutTx1.join();|Tx2 waits on attempt to acquire x-lock.
*Note:* works correctly with reverse transaction order.|correct|X-lock for tx2
fails, as tx2 younger than tx1 and is not allowed to wait for tx1, it should
DIE instead.|
|2|*testWaitDie1*
var tx1 = beginTx();var tx2 = beginTx();
var key1 = key("test");
xlock(tx2, key1).join();
CompletableFuture<?> xlockFutTx1 = xlock(tx1, key1);
{_}assertFalse{_}(xlockFutTx1.isDone());
commitTx(tx2);
xlockFutTx1.join();|Tx1 fails on attempt of acquiring the x-lock
*Note:* works correctly with reverse transaction order.|correct|tx2
successfully acquires x-lock.
tx1 waits on attempt of acquiring x-lock.
After tx2 is committed, the future for tx1 is completed without exception.|
was (Author: denis chudov):
*Correct* means that the behavior matches the expectations.
||N||Scenario||Ignite 3||MVP||Expected||
|1|*testWaitDie0*
var tx1 = beginTx();
var tx2 = beginTx();
var key1 = key("test");
xlock(tx2, key1).join();
CompletableFuture<?> xlockFutTx1 = xlock(tx1, key1);
assertFalse(xlockFutTx1.isDone());
commitTx(tx2);
xlockFutTx1.join();|Tx2 waits on attempt to acquire x-lock.
*Note:* works correctly with reverse transaction order.|correct|X-lock for tx2
fails, as tx2 younger than tx1 and is not allowed to wait for tx1, it should
DIE instead.|
|2|*testWaitDie1*
*var tx1 = beginTx();var tx2 = beginTx();
*
*var key1 = key("test");
*
*xlock(tx2, key1).join();*
*CompletableFuture<?> xlockFutTx1 = xlock(tx1, key1);
{_}assertFalse{_}(xlockFutTx1.isDone());
*
*commitTx(tx2);
xlockFutTx1.join();*|Tx1 fails on attempt of acquiring the x-lock
*Note:* works correctly with reverse transaction order.|correct|tx2
successfully acquires x-lock.
tx1 waits on attempt of acquiring x-lock.
After tx2 is committed, the future for tx1 is completed without exception.|
> Compare deadlock prevention implementations and work out decisions about
> correct behavior in corner cases
> ---------------------------------------------------------------------------------------------------------
>
> Key: IGNITE-18091
> URL: https://issues.apache.org/jira/browse/IGNITE-18091
> Project: Ignite
> Issue Type: Task
> Reporter: Denis Chudov
> Assignee: Denis Chudov
> Priority: Major
> Labels: ignite-3
>
> *Motivation*
> Today we have several possible implementations of deadlock prevention: AI 3
> [1], transactions POC [2], and possible implementation based on concurrency
> control paper [3]. Moreover, we have a multiple granularity lock model [4]
> where some lock modes are compatible, others are not, which allows
> reenterability in some cases, and sharing of locks between transactions. We
> should understand the differences of behavior of each implementation in
> different scenarios, and how it matches with our expectations.
> *Definition of done*
> Table with a set of scenarios and description of behavior, including the
> expected one.
> [1] org.apache.ignite.internal.tx.impl.HeapLockManager
> [2] https://github.com/ascherbakoff/ai3-txn-mvp
> [3] https://dl.acm.org/doi/pdf/10.1145/320251.320260
> [4] https://web.stanford.edu/class/cs245/readings/granularity-of-locks.pdf
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
