[
https://issues.apache.org/jira/browse/RANGER-5201?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Naoki Takezoe updated RANGER-5201:
----------------------------------
Summary: Retry GET requests on HTTP 5xx errors in RangerRESTClient (was:
Retry GET request on server error in RangerRESTClient)
> Retry GET requests on HTTP 5xx errors in RangerRESTClient
> ---------------------------------------------------------
>
> Key: RANGER-5201
> URL: https://issues.apache.org/jira/browse/RANGER-5201
> Project: Ranger
> Issue Type: Improvement
> Components: plugins
> Reporter: Naoki Takezoe
> Priority: Minor
> Time Spent: 10m
> Remaining Estimate: 0h
>
> RangerRESTClient has a retry mechanism for connectivity errors that was
> implemented in https://issues.apache.org/jira/browse/RANGER-3565 but it
> doesn't retry HTTP 5xx errors because it works only when
> ClientHandlerException is thrown:
> https://github.com/apache/ranger/blob/dbaad69de10ff7b21b5bd3fc08c4b65b4aa25dff/agents-common/src/main/java/org/apache/ranger/plugin/util/RangerRESTClient.java#L409-L415
> while Jersey's default client handler that is used in RangerRESTClient
> doesn't raise an exception even for 5xx errors:
> https://github.com/javaee/jersey-1.x/blob/1.19.3/jersey-client/src/main/java/com/sun/jersey/client/urlconnection/URLConnectionClientHandler.java
> Therefore, policy retrieval immediately fails without retry when, for
> example, ranger-admin behind a reverse proxy is temporarily unavailable.
> It's better to retry GET requests also on HTTP 5xx errors in RangerRESTClient.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
