[ 
https://issues.apache.org/jira/browse/KUDU-1549?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15409923#comment-15409923
 ] 

Adar Dembo commented on KUDU-1549:
----------------------------------

Some interesting factoids about this tserver:
- 391 log block containers whose last records were truncated
- Each container has about 3 MB of data, total of 1 GB. This isn't the total 
amount of data managed by the tserver; we only log the containers whose last 
record was truncated. 
- It took about 22m to load the filesystem.
- After that, there were 703 tablets registered.

[~bruceSz], how much total data is under management? In 
/export/servers/kudu/tserver_data_7052/data, how many .data files are there? 
What is their total size? Likewise for .metadata files. My suspicion is that, 
given the number of tablets and the size of each block, we may have so many 
blocks that the metadata files are quite large and it takes some time to read 
them all. Note that we don't actually read all the data on startup; we just 
read the metadata.

Separately, there's an interesting issue going on with tablet 
a14844513e5243a993b2b84bf0dcec4c. It appears to be caught in a loop where it 
remote bootstraps itself, then is tombstoned. This happens over ~8000 times, 
and it constitutes almost half of the logged lines. I'll try to attach a log 
with references to just that tablet.


> recovery speed of kudu-tserver should be faster.
> ------------------------------------------------
>
>                 Key: KUDU-1549
>                 URL: https://issues.apache.org/jira/browse/KUDU-1549
>             Project: Kudu
>          Issue Type: Improvement
>         Environment: cpu: Intel(R) Xeon(R) CPU E5-2660 v3 @ 2.60GHz
> mem: 252 G
> disk: single ssd  1.5 T left.
>            Reporter: zhangsong
>
> After experiencing physical node crash, it found recovery/start speed of 
> kudu-tserver is slower than that of usual restart case.  There are some 
> message like "Found partial trailing metadata" in kudu-tserver log and it 
> seems cost more than 20 minute to recover these metadata.
> According to adar , it should be this slow.
> attachment is the start log .



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to