266948 - streaming bi-directional test mode for selfserv and strsclnt

Status: NEW

(NSS :: Libraries, enhancement, P2)

Description

[Julien Pierre]

Currently, strsclnt can only make a single short GET HTTP request to the server.
There is no way to make a POST request or do anything else that can send a lot
of data to the server.

We need a way to test client to server throughput and bi-directional throughput.
The simplest way to do this might be to have them talk something other than
HTTP, ie. a simple byte protocol where each side sends a header with the # of
bytes to follow. This may however require using non-blocking sockets to get
right if only one side is sending data to avoid deadlocks; or multiple threads
per socket (one for each side) with the full-duplex option set, and of course
proper synchronization for the close case.

Comment 1

[Julien Pierre]

Nelson mentioned that the -2 option already has some of the code we need to
implement this.

Comment 2

[Julien Pierre]

Attachment: strsclnt patch

This patch is an experiment to attempt to do bidirectional I/O with strsclnt .
It also tries to fix the issue where a thread is restarted for each request.
The patch has bugs and hangs have been observed with it. But it is a starting
point.

Comment 3

[Julien Pierre]

Attachment: matching selfserv patch

Goes together with the strsclnt patch. Also broken .

Comment 4

[Julien Pierre]

Alexei, this is the bug we were discussing earlier today.
The patch to strsclnt will need to be merged since the threading fixes are part
of bugzilla 292151 and will be checked in soon.

Right now, the logic contained in these patches to do the HTTP POST request in
strsclnt and send large responses in selfserv doesn't work properly. A long
stress test will result in all the threads eventually hanging on both sides.

Comment 5

[Julien Pierre]

When this is implemented, there should be a way to configure the size of the
requests and responses, ie. HTTP content-length . I would suggest adding a
command-line argument for both strsclnt and selfserv.

The HTTP content-length header wouldn't necessarily have to be parsed on either
the client or server side. The client could just send any data it wants and then
call PR_Shutdown() to shut down the send side of the connection. The server
could PR_Read until it gets an EOF (0), then start sending, then PR_Close. The
client would read the response until EOF . This is what I was trying accomplish
in the patches I attached, but something was wrong and needs to be fixed. I have
implemented other clients with this approach before, and it worked fine.

Comment 6

[Nelson Bolyard (seldom reads bugmail)]

Unsetting target milestone in unresolved bugs whose targets have passed.

Comment 7

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee is inactive on Bugzilla, and this bug has priority 'P2'.
:beurdouche, could you have a look please?

For more information, please visit auto_nag documentation.

Comment 8

[Marco Castelluccio [:marco]]

We have modified the bot to only consider P1 as high priority, so I'm cancelling the needinfo here.