- The net code no longer has a recv_queue, instead reuse same recv buffer.
- Introduce net_recv_reset() to reset the recv buffer when needed.
- Have the workers spread the load better between them by slightly
delaying their next accept lock and giving them an accept treshold
so they don't go ahead and keep accepting connections if they end
up winning the race constantly between the workers.
- The kore_worker_acceptlock_release() is no longer available.
- Prepopulate the HTTP server response header that is added to each
response in both normal HTTP and SPDY modes.
- The path and host members of http_request are now allocated on the heap.
These changes overall result better performance on a multicore machine,
especially the worker load changes shine through.
This commit renames certain POST centric variable and configuration
naming to the correct HTTP body stuff.
API changes include http_postbody_text() and http_postbody_bytes() to
have become http_body_text() and http_body_bytes().
The developer is still responsible for validating the method their
page handler is called with. Hopefully this becomes a configuration
option soon enough.
Instead of letting http_requests spin, if we cannot allocate
a connection for the request we will queue them up put them to sleep.
When a connection becomes available, we'll wake up a request that
was waiting for a connection and let it continue.
This completely avoids consuming massive amounts of cpu time
when dealing with thousands of requests waiting for a pgsql
worker to become ready.
Gone is the ugly KORE_PGSQL macro that hid an overly complex
state machine for the pgsql api.
Gone is the pgsql array that was attached to http_requests.
Gone are the callback hacks inside the pgsql api.
Instead, I strongly encourage people to use the new state machine
api Kore offers to properly deal with asynchronous queries.
The pgsql example in examples/pgsql has been updated to reflect
these changes.
In order to use this, define states for your page handler:
struct http_state mystates[] = {
{ "PAGE_STATE_INIT", page_init },
{ "PAGE_STATE_RESULT", page_result },
};
In your page handler you can then simply call http_state_run() with
your states and http_request. This will cause Kore to start calling
your state callbacks beginning at index 0.
State callbacks have the same prototype as page handlers:
int func(struct http_request *);
However, unlike page handlers they MUST return one of the following:
- HTTP_STATE_OK: All good, just continue the fsm.
- HTTP_STATE_ERROR: Abort fsm and return KORE_RESULT_OK to Kore
(This will cancel the http request).
- HTTP_STATE_RETRY: Return KORE_RESULT_RETRY to Kore.
(Kore will retry your page handler next event loop).
- HTTP_STATE_COMPLETE: The fsm completed, break out cleanly.
Note that using this is completely optional and you can still
use the traditional way of writing page handlers.
The fsm is designed to get rid of the clutter that exists today
in Kore when dealing with non blocking tasks or pgsql calls.
This way we can get our code called whenever a stream is
completed. This cb handler does stand alone from an http_request
and is passed a netbuf data structure.
* Always make sure we end the stream properly
* Check for SPDY_FLOW_WINDOW_MAX on window frame updates
* Kill SPDY_STREAM_BLOCKING, once flow control kicks in its per session
These 2 functions can be used to move an HTTP request
from/to the active http_requests list. Effectively
putting them to "sleep" or "waking them up".
Sprinkle this through the pgsql and task code.
If used correctly greatly reduces overhead for
managing sleeping tasks.
Synchronize access to state/result properly so one
can access these from inside the task as well.
Introduce KORE_TASK_STATE_ABORT which will be set
when a task needs to be abort. You can use this
to create tasks that run in a loop until aborted.
Instead if a task is used from inside a request
you MUST call kore_task_bind_request() on it.
This way we can move forward for tasks that
don't belong to page handlers.
Also, some bug fixes for removing http_requests
that are indeed linked to a currently running task.
Tasks are now assigned to available threads instead
of a global task list.
You can now pass messages between your page handler
and the created task using the kore_task_channel_*
functions.
Only one task per time can be assigned to a request
but I feel this is probably a bad design choice.
Preferably we'd want to be able to start tasks
regardless of being in a page handler or not,
this not only ads flexibility but seems like
a better choice overall as it opens a lot more
possibilities about how tasks can be used.
Don't wait for a full event loop until we call the page handler
for a received pgsql result. This speeds up page loads using
KORE_PGSQL by quite a lot, especially on a non busy server.
Including but not limited to:
- Correctly use PQerrorMessage() in case we cleanup with PQfinish
- If we get a network error, cleanup the connection
- No longer call the page handler from inside kore_pgsql_handle()
but instead just put it to sleep in case we don't need it.
This does grow the http_requests list quite a bit with sleeping
connections and can perhaps be improved later on.
- Allow us to on error return OK from a page handler from inside
the completetion block for KORE_PGSQL().
- Count the cummulative time for a request to finish instead
of the latest run time for the handler.
Has support for full async pgsql queries. Most of the logic
is hidden behind a KORE_PGSQL() macro allowing you to insert
these pgsql calls in your page handlers without blocking the
kore worker while the query is going off.
There is place for improvement here, and perhaps KORE_PGSQL won't
stay as I feel this might overcomplicate things instead of making
them simpler as I thought it would.
Joris Vink