[
https://issues.apache.org/jira/browse/HADOOP-17276?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17202490#comment-17202490
]
Mingliang Liu edited comment on HADOOP-17276 at 9/26/20, 2:07 AM:
------------------------------------------------------------------
So, for the code reviews, I'll leave comments about implementation in the pull
request. Allow me sort my understanding out here:
# This is compatible change because it does not enforce any format of the
existing caller context. This is very important to confirm. The initial
{{CallerContext}} design used a plain string format instead of
{{key:value,key:value}} format for sake of flexibility and simplicity. Audit
log per se is using {{key=value}} format and was not designed with nested
structure in mind. I believe every time we revisit audit format, people would
think of better ways e.g. JSON. But compatibility - not only code setting it
but also existing tools parsing it - is always one of the key concern.
# Here for easier appending more information into the caller context in an
organized way, in this JIRA we propose that user can call multiple times
{{builder.append(key, value)}} to append information into the context. The
CallContext object once constructed is still immutable as it is. So we are not
changing any existing behavior in this proposal.
Given above points, I would be +1 on this.
For the implementation, one point is not allowing "=" or "\t" as the separator
of the key/value items in the context string. This is because the audit log
itself is using = to separate key/values in every top-level field, and using
"\t" to separate top-level fields.
CC: [~jitendra] [~stevel] [~inigoiri] and [~aajisaka]
Thanks,
[EDIT]: tagged wrong Goiri user, fixed
was (Author: liuml07):
So, for the code reviews, I'll leave comments about implementation in the pull
request. Allow me sort my understanding out here:
# This is compatible change because it does not enforce any format of the
existing caller context. This is very important to confirm. The initial
{{CallerContext}} design used a plain string format instead of
{{key:value,key:value}} format for sake of flexibility and simplicity. Audit
log per se is using {{key=value}} format and was not designed with nested
structure in mind. I believe every time we revisit audit format, people would
think of better ways e.g. JSON. But compatibility - not only code setting it
but also existing tools parsing it - is always one of the key concern.
# Here for easier appending more information into the caller context in an
organized way, in this JIRA we propose that user can call multiple times
{{builder.append(key, value)}} to append information into the context. The
CallContext object once constructed is still immutable as it is. So we are not
changing any existing behavior in this proposal.
Given above points, I would be +1 on this.
For the implementation, one point is not allowing "=" or "\t" as the separator
of the key/value items in the context string. This is because the audit log
itself is using = to separate key/values in every top-level field, and using
"\t" to separate top-level fields.
CC: [~jitendra] [~stevel] [~inigoiri] and [~aajisaka]
Thanks,
> Extend CallerContext to make it include many items
> --------------------------------------------------
>
> Key: HADOOP-17276
> URL: https://issues.apache.org/jira/browse/HADOOP-17276
> Project: Hadoop Common
> Issue Type: Improvement
> Reporter: Hui Fei
> Assignee: Hui Fei
> Priority: Major
> Labels: pull-request-available
> Time Spent: 2h 50m
> Remaining Estimate: 0h
>
> Now context is string. We need to extend the CallerContext because context
> may contains many items.
> Items include
> * router ip
> * MR or CLI
> * etc
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
