Avery Pennarun
ca14231aae
Whoops, we were spinning in the server when the mux outbuf was full.
...
Once again, the buffering gets mixed up with the selecting. Seems to be the
story of my life.
2010-05-02 06:17:43 -04:00
Avery Pennarun
6c2dc54b9e
More latency fixes: use a PING/PONG pair to limit queue length.
...
It seems ssh is kind of stupid and uses a really big SO_SNDBUF (hundreds of
kbytes). Thus, we can't depend on the socket's output buffer to limit our
latency down to something reasonable. Instead, limit the amount of data we
can send in a single round trip.
2010-05-02 05:39:17 -04:00
Avery Pennarun
5d1390927d
Don't overfill the mux's output buffer.
...
Otherwise a single busy stream can ruin it for everybody.
2010-05-02 05:06:51 -04:00
Avery Pennarun
81c89ce9be
Don't bother with a backtrace when we produce certain fatal errors.
...
We'll introduce a new "Fatal" exception for this purpose, and throw it when
we just want to print a user message and abort immediately.
2010-05-02 02:29:23 -04:00
Avery Pennarun
2dd328ada4
Add a -v (and -vv) flag and decrease default message verbosity.
2010-05-02 02:14:20 -04:00
Avery Pennarun
bfd506dcdc
Improve some debugging information to find the weird data problem.
...
Turns out list.pop() removes the *last* item, not the first one. Oops. It
all works great for queues of only one item... :)
2010-05-02 01:25:09 -04:00
Avery Pennarun
915a96b0ec
We now have a server that works... some of the time.
...
There still seem to be some weird timing and/or closing-related bugs, since
I can't load the eqldata project correctly unless I use --noserver.
2010-05-02 00:52:06 -04:00
