[ 
https://issues.apache.org/jira/browse/IGNITE-9903?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16755142#comment-16755142
 ] 

Andrey Gura edited comment on IGNITE-9903 at 1/29/19 6:40 PM:
--------------------------------------------------------------

[~astelmak] I've looked at your change and have some comments:

- Size of {{SWITCH_SEGMENT_RECORD}} should be calculated using 
{{RecordSerializer.size}} (see example here: 
{{org.apache.ignite.internal.processors.cache.persistence.wal.filehandle.FileWriteHandleImpl#close}}).
- Unused import in all changed files.
- {{switchSegmentRecordOffset}} array never updates (except of {{0}} value). 
Please add test for this situation.

Could you please fix this comments? Thanks.



was (Author: agura):
[~astelmak] I've looked at your change and have some comments:

- Size of {{SWITCH_SEGMENT_RECORD}} should be calculated using 
{{RecordSerializer.size}} (see example here: 
{{org.apache.ignite.internal.processors.cache.persistence.wal.filehandle.FileWriteHandleImpl#close}}).
- Unused import in all changed files.
- {{switchSegmentRecordOffset}} array never updates (except of {{0}} value).

Could you please fix this comments? Thanks.


> Copy only actual amount of data during archiving of WAL segment
> ---------------------------------------------------------------
>
>                 Key: IGNITE-9903
>                 URL: https://issues.apache.org/jira/browse/IGNITE-9903
>             Project: Ignite
>          Issue Type: Improvement
>            Reporter: Andrey Gura
>            Assignee: Alexey Stelmak
>            Priority: Major
>             Fix For: 2.8
>
>
> Current WAL archiver implementation copies full WAL segment to the archive 
> while actual size of the segment can be significantly less then 
> {{maxWalSegmentSize}} (segment files are preallocated for max possible 
> segment size). 
> In order to reduce disk space consuming we need copy only actual data of 
> segment using {{SWITCH_SEGMENT_RECORD}} as marker. It will require some kind 
> of WAL iterator implementation that will just copy WAL records using 
> information about record size without records deserialization.



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

Reply via email to