[
https://issues.apache.org/jira/browse/IGNITE-18045?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vladislav Pyatkov updated IGNITE-18045:
---------------------------------------
Description:
[Transaction
IEP|https://cwiki.apache.org/confluence/display/IGNITE/IEP-91%3A+Transaction+protocol#IEP91:Transactionprotocol-Uniqueindex]
contains an operation that need a short term locks (look at insert operation
for sorted index), but this type of lock lies on present implementation badly.
Today, if a transaction releases a lock that has been got to twice (maybe
different types, but in one key) the lock is released at all.
Adding counters by lock types is solved the issue, but possible it is not
required to solve the problem, because we can change operation in order of the
short term locks will be not needed.
was:
[Transaction
IEP|https://cwiki.apache.org/confluence/display/IGNITE/IEP-91%3A+Transaction+protocol#IEP91:Transactionprotocol-Uniqueindex]
contains an operation that need a short term locks, but this type of lock lies
on present implementation badly. Today, if a transaction releases a lock that
has been got to twice (maybe different types, but in one key) the lock is
released at all.
Adding counters by lock types is solved the issue, but possible it is not
required to solve the problem, because we can change operation in order of the
short term locks will be not needed.
> Short term locks support
> ------------------------
>
> Key: IGNITE-18045
> URL: https://issues.apache.org/jira/browse/IGNITE-18045
> Project: Ignite
> Issue Type: Improvement
> Reporter: Vladislav Pyatkov
> Priority: Major
> Labels: ignite-3
>
> [Transaction
> IEP|https://cwiki.apache.org/confluence/display/IGNITE/IEP-91%3A+Transaction+protocol#IEP91:Transactionprotocol-Uniqueindex]
> contains an operation that need a short term locks (look at insert operation
> for sorted index), but this type of lock lies on present implementation
> badly. Today, if a transaction releases a lock that has been got to twice
> (maybe different types, but in one key) the lock is released at all.
> Adding counters by lock types is solved the issue, but possible it is not
> required to solve the problem, because we can change operation in order of
> the short term locks will be not needed.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
