[
https://issues.apache.org/jira/browse/ZOOKEEPER-4743?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
HaiyuanZhao updated ZOOKEEPER-4743:
-----------------------------------
Description:
DataVersion is a 32-bit signed integer. When dataVersion exceeds Integer.MAX,
an overflow problem will occur.
According to my own understanding, overflow itself is not a big problem. The
core of dataVersion is to describe the causal order. Two different values can
be used to determine the happened-before relationship between the two.
Then even if an overflow occurs, except for the scenario where Integer.MAX is
larger than Integer.MAX+1, compared with any value in other scenarios and the
value + 1, the latter is always larger than the former. As described above for
the role of dataVersion, if it continues to increase after overflow, the role
will still take effect.
It is also worth nothing that zookeeper has special processing for
dataVersion=-1. When expectedVersion==and currentVersion==-1, the dataVersion
equality check is skipped, which violates the semantics of CAS updates. As
shown in the following code block
{code:java}
private static int checkAndIncVersion(int currentVersion, int
expectedVersion, String path) throws KeeperException.BadVersionException {
if (expectedVersion != -1 && expectedVersion != currentVersion) {
throw new KeeperException.BadVersionException(path);
}
return currentVersion + 1;
}
{code}
I propose that when the current dataVersion is -2, update dataVersion to 0 to
skip the case of -1(Inspired by [~Tison],). This will avoid breaking the
semantics of CAS updates.
was:
DataVersion is a 32-bit signed integer. When dataVersion exceeds Integer.MAX,
an overflow problem will occur.
According to my own understanding, overflow itself is not a big problem. The
core of dataVersion is to describe the causal order. Two different values can
be used to determine the happened-before relationship between the two.
Then even if an overflow occurs, except for the scenario where Integer.MAX is
larger than Integer.MAX+1, compared with any value in other scenarios and the
value + 1, the latter is always larger than the former. As described above for
the role of dataVersion, if it continues to increase after overflow, the role
will still take effect.
It is also worth noting that zookeeper has special processing for
dataVersion=-1. When expectedVersion==and currentVersion==-1, the dataVersion
equality check is skipped, which violates the semantics of CAS updates. As
shown in the following code block
{code:java}
private static int checkAndIncVersion(int currentVersion, int
expectedVersion, String path) throws KeeperException.BadVersionException {
if (expectedVersion != -1 && expectedVersion != currentVersion) {
throw new KeeperException.BadVersionException(path);
}
return currentVersion + 1;
}
{code}
I propose that when the current dataVersion is -2, update dataVersion to 0 to
skip the case of -1(Inspired by [~Tison],). This will avoid breaking the
semantics of CAS updates.
> Jump from -2 to 0 when zookeeper increments znode dataVersion
> -------------------------------------------------------------
>
> Key: ZOOKEEPER-4743
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-4743
> Project: ZooKeeper
> Issue Type: Improvement
> Reporter: HaiyuanZhao
> Priority: Major
> Labels: pull-request-available
> Time Spent: 20m
> Remaining Estimate: 0h
>
> DataVersion is a 32-bit signed integer. When dataVersion exceeds Integer.MAX,
> an overflow problem will occur.
> According to my own understanding, overflow itself is not a big problem. The
> core of dataVersion is to describe the causal order. Two different values
> can be used to determine the happened-before relationship between the two.
> Then even if an overflow occurs, except for the scenario where Integer.MAX is
> larger than Integer.MAX+1, compared with any value in other scenarios and the
> value + 1, the latter is always larger than the former. As described above
> for the role of dataVersion, if it continues to increase after overflow, the
> role will still take effect.
> It is also worth nothing that zookeeper has special processing for
> dataVersion=-1. When expectedVersion==and currentVersion==-1, the dataVersion
> equality check is skipped, which violates the semantics of CAS updates. As
> shown in the following code block
> {code:java}
> private static int checkAndIncVersion(int currentVersion, int
> expectedVersion, String path) throws KeeperException.BadVersionException {
> if (expectedVersion != -1 && expectedVersion != currentVersion) {
> throw new KeeperException.BadVersionException(path);
> }
> return currentVersion + 1;
> }
> {code}
> I propose that when the current dataVersion is -2, update dataVersion to 0 to
> skip the case of -1(Inspired by [~Tison],). This will avoid breaking the
> semantics of CAS updates.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
