Refactor listen/notify as a pubsub backend #6811
Conversation
pedro-psb
force-pushed
the
tasking/add-pubsub-interface
branch
from
d976358 to
346574b
Compare
pedro-psb
force-pushed
the
tasking/add-pubsub-interface
branch
3 times, most recently
from
f25b950 to
6a9000d
Compare
| self.sentinel_r, self.sentinel_w = os.pipe() | ||
| os.set_blocking(self.sentinel_r, False) | ||
| os.set_blocking(self.sentinel_w, False) |
There was a problem hiding this comment.
Is this really related? I thought this os.pipe was primarily part of the select calls to receive the signals like SIGINT, SIGTERM.
There was a problem hiding this comment.
I added this to make the semantics of the PubSub readiness tied to the data it actually stores, which is the self.message_buffers queue (not the connection buffer). If I just use connection.connection, test_pubsub:111 fails as the pg_notify_callback is called automatically on same-process connections and is empty at this point.
But I don't feel very good about how complex this looks. I think using notifies may make our lives easier
pedro-psb
force-pushed
the
tasking/add-pubsub-interface
branch
4 times, most recently
from
aea9033 to
5e71216
Compare
pedro-psb
force-pushed
the
tasking/add-pubsub-interface
branch
2 times, most recently
from
6389b3a to
a04fa6a
Compare
|
Here is one of the general ideas for why I prefer to not add a callback into the pubsub interface (the other is testability). 
|
pedro-psb
force-pushed
the
tasking/add-pubsub-interface
branch
from
a04fa6a to
a508837
Compare
This reverts commit 6e6879c.
pedro-psb
force-pushed
the
tasking/add-pubsub-interface
branch
from
a508837 to
cecf943
Compare
No description provided.