[
https://issues.apache.org/jira/browse/ATLAS-602?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15266396#comment-15266396
]
Shwetha G S commented on ATLAS-602:
-----------------------------------
Each hook invocation does full entity update. So, the first command will
create/update tmp_table entity with columns a,b. Second command will
create/update tmp_table entity with columns a,b,c. Even if the first command is
lost, the 2nd command will create the latest state of the entity. So, I think
its ok to log and return from AtlasHook.notifyEntities() after retries.
There can be some edge cases like:
1. create table t1
2. drop table t1
3. create table t1
If 2nd command here is lost, 3rd command will update the existing entity
instead of creating new entity. We need to process the failed messages
manually. But its better than blocking a thread.
> Hooks stuck in case of failure
> ------------------------------
>
> Key: ATLAS-602
> URL: https://issues.apache.org/jira/browse/ATLAS-602
> Project: Atlas
> Issue Type: Bug
> Reporter: Shwetha G S
> Assignee: Vimal Sharma
> Priority: Critical
> Fix For: 0.7-incubating
>
>
> in the case of whoever uses AtlasHook.notifyEntities(), the retry loop
> doesn't ever exit, which is a problem as well. Thanks [~yhemanth] for
> reporting it
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
