[jira] [Commented] (ASTERIXDB-2444) Potential data loss due to clock drift

["ASF subversion and git services (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ASTERIXDB-2444?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16612507#comment-16612507 ] 

ASF subversion and git services commented on ASTERIXDB-2444:
------------------------------------------------------------

Commit 5550a5f225a0528fe2229d2b39db4595590ab0ea in asterixdb's branch refs/heads/master from [~mhubail]
[ https://git-wip-us.apache.org/repos/asf?p=asterixdb.git;h=5550a5f ]

[ASTERIXDB-2444][STO] Avoid Using System Clock in Checkpoints

- user model changes: no
- storage format changes: yes
- interface changes: yes

Details:
- Replace the usage of system clock timestamps in checkpoints
  by a sequencer.
- Update Asterix/Hyracks storage version to reflect the recent
  changes in storage.
- This change is expected to break storage backward compatibility.

Change-Id: Idc061e6eaccfb308b29a5a263b77a0a849694d4f
Reviewed-on: https://asterix-gerrit.ics.uci.edu/2961
Reviewed-by: Michael Blow <mb...@apache.org>
Integration-Tests: Michael Blow <mb...@apache.org>
Tested-by: Michael Blow <mb...@apache.org>


> Potential data loss due to clock drift
> --------------------------------------
>
>                 Key: ASTERIXDB-2444
>                 URL: https://issues.apache.org/jira/browse/ASTERIXDB-2444
>             Project: Apache AsterixDB
>          Issue Type: Bug
>          Components: STO - Storage
>    Affects Versions: 0.9.4
>            Reporter: Murtadha Hubail
>            Assignee: Murtadha Hubail
>            Priority: Major
>             Fix For: 0.9.5
>
>
> Currently we use a timestamp based on the current system time to sort the index files to identify the order in which the index components should be searched. If the system clock is changed to a time earlier than the latest index component time and a new component written, it will result in sorting the components in a wrong order. This issue can also be encountered if a current master replica has a system time that is ahead of another replica. If the current master fails and that replica is promoted, it might generate new components with an earlier timestamp. The same issue is applicable in component ids.
> Proposed Fix:
> Avoid using the system clock timestamp and use a simple monotonically increasing sequencer.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)