[
https://issues.apache.org/jira/browse/HBASE-20826?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16528419#comment-16528419
]
Ted Yu commented on HBASE-20826:
--------------------------------
w.r.t. Sergey's comment, here is what I think:
1. without actually generating the full output from
{{ProtobufUtil.getShortTextFormat}}, we would want to know how long the output
would be.
How about adding a new method with limit of length:
{code}
public static String getShortTextFormat(Message m, int lengthLimit) {
{code}
The method would return midway if limit is exceeded. I haven't gone thru all
the Request types involved in the method so there may be some difficulty
implementing the new method.
2. w.r.t. not truncating, I kind of agree. How about selectively truncating
long output ? Meaning, we only show complete output for N% of each request type
where output is long (N is low) ?
> Truncate responseInfo attributes on RpcServer WARN messages
> -----------------------------------------------------------
>
> Key: HBASE-20826
> URL: https://issues.apache.org/jira/browse/HBASE-20826
> Project: HBase
> Issue Type: Improvement
> Components: rpc
> Reporter: Sergey Soldatov
> Assignee: Josh Elser
> Priority: Major
> Fix For: 3.0.0, 2.1.0, 2.0.2, 2.2.0
>
> Attachments: HBASE-20826.001.branch-2.0.patch,
> HBASE-20826.002.branch-2.0.patch
>
>
> With Phoenix in the picture, dumping the {{Call}} protobuf to the RS log can
> get *really* chatty, real fast. Notably, some serialized filters just spam
> the log with binary garbage.
> Let's add an upper-limit to the length of params we'll put out at WARN, and
> leave the full content for TRACE.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
