Plumb message_queue_nonempty callback #164
Conversation
|
@quodlibetor or @thijsc, can I talk one of you into reviewing this? |
|
Wouldn't it be possible to use a callback on the future? Or does that have downsides for you? |
|
Yeah, at @MaterializeInc we're not using the futures consumer. We have our own event loop that calls |
|
Come to think of it, it should be possible to avoid the separate thread in the StreamConsumer entirely with the functionality in this PR. I'm going to poke at that a bit. |
This makes some timeout logic in a forthcoming PR much clearer. I think
it also admits a clearer API, as
consumer.poll(Timeout::Never)
is quite a bit more readable than
consumer.poll(None).
Note that this is not actually a breaking API change, as
`From<Duration>` and `From<Option<Duration>>` are properly implemented
for `Timeout`, and so `consumer.poll(None)`, `consumer.poll(duration)`,
and `consumer.poll(Some(duration))` all compile as before.
There is one small API change to the admin client. Previously it was not
possible to use the builder API to unset a request timeout or operation
timeout; now it is. (This was an oversight on my part; the API should
previously have taken Option<Option<Duration>>, though that's quite a
mouthful.) Luckily, since the admin client has not yet made it into a
release, we're within rights to change the API.
This makes it possible to receive more targeted notifications when there is activity on just one queue.
benesch
force-pushed
the
ready-callbacks
branch
from
24a5a71 to
b8ef316
Compare
This makes it possible to receive a notification via a Rust callback when there are new messages available.
benesch
force-pushed
the
ready-callbacks
branch
from
b8ef316 to
8822bbf
Compare
This turns out to be fairly difficult, so I'm just going to merge as is, now that tests are passing. (Just needed to increase the timeout for CI, because things seem to be a bit slower due to the small Travis VMs.) |
This makes it possible to receive a notification via a Rust callback when there are new messages available. The bulk of the implementation is in the last two commits; the first commit is just preparatory work.