[
https://issues.apache.org/jira/browse/TS-4321?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15242110#comment-15242110
]
ASF GitHub Bot commented on TS-4321:
------------------------------------
Github user maskit commented on the pull request:
https://github.com/apache/trafficserver/pull/570#issuecomment-210199731
Reverting TS-4321 sounds reasonable if we can modify
``stream->response_header`` without any side effects.
The response headers in FetchSM must not be modified because FetchSM uses
them internally to check if the response body is chunked. It seems
``Http2Stream::response_initialize_data_handling`` does the similar thing.
So all checks depend on response headers must be done before we modify
headers here if we don't make the copy. If we cannot assure that, we still need
the fix for TS-4321.
Moreover, generating H2 headers from headers in Http2Stream sounds odd.
> H2 truncates certain contents on master
> ---------------------------------------
>
> Key: TS-4321
> URL: https://issues.apache.org/jira/browse/TS-4321
> Project: Traffic Server
> Issue Type: Bug
> Components: HTTP/2
> Reporter: Leif Hedstrom
> Assignee: Masakazu Kitajo
> Priority: Critical
> Labels: A, regression
> Fix For: 6.2.0
>
>
> This seems to be a regression, since it does not happen on 6.1.x. I haven't
> dug deeper into it, but in the case I found, ATS closes the stream after 2
> data frames (8k + a few bytes), where it really needs to send about 27KB.
> I've reverted my server to 6.1.x for now, but I'll see if I can find
> something to reproduce it on e.g. docs.trafficserver.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
