Couchbase Server

memory spike due to growing number of replica checkpoints if some of the nodes have slow disk write

Details

Description

a solution proposed for this issue is to coalesce checkpoints in the replica side if number of open checkpoints is more than two

https://github.com/membase/ep-engine/commit/dd4d60b79008586247e86c62e9ad2cc9320867d9

fixed in branch-17 and is going to be given to users as hotfix as need basis

Activity

Descending order - Click to sort in ascending order
Hide
Permalink
Chiyoung Seo added a comment -
We did several enhancements on the checkpoint datastructure to reduce the memory overhead.

1) Collapse multiple checkpoints into a single one
2) Each checkpoint item does not point to any Blob value instance anymore. Instead, it will always gets the latest value from memory hash table.
3) Adapt the checkpoint item datastructure and reduce its memory overhead from 88 bytes to 40 bytes.
Show
Chiyoung Seo added a comment - We did several enhancements on the checkpoint datastructure to reduce the memory overhead. 1) Collapse multiple checkpoints into a single one 2) Each checkpoint item does not point to any Blob value instance anymore. Instead, it will always gets the latest value from memory hash table. 3) Adapt the checkpoint item datastructure and reduce its memory overhead from 88 bytes to 40 bytes.
Hide
Permalink
Chiyoung Seo added a comment -
It was already merged into 1.8 branch. The last 1.8 build that the QA team has been testing includes this fix.
Show
Chiyoung Seo added a comment - It was already merged into 1.8 branch. The last 1.8 build that the QA team has been testing includes this fix.
Hide
Permalink
Farshid Ghods added a comment -
fixed but not merged yet to 1.8 branch
Show
Farshid Ghods added a comment - fixed but not merged yet to 1.8 branch

People

  • Assignee:
    Chiyoung Seo
    Reporter:
    Farshid Ghods
Vote (0)
Watch (2)

Dates

  • Created:
    Updated:
    Resolved: