158 Commits

Author	SHA1	Message	Date
Cory McWilliams	6af36cafa9	ssb: Wait, I don't think we need to close any connections here.	2025-06-25 20:10:22 -04:00
Cory McWilliams	fca859d93d	ssb: Are we inadvertantly closing connections when an inner tunnel request ends?	2025-06-25 20:00:44 -04:00
Cory McWilliams	53cba2d7e4	Move all but the deleting off of the writer when deleting blobs and messages. Trying to eliminate a period of unresponsiveness near launch.	2025-06-15 08:46:07 -04:00
Cory McWilliams	a1f1eb34d5	ssb: sequence: Sequential 32-bit integer starting at 1.	2025-06-11 20:12:23 -04:00
Cory McWilliams	2f0a2ac6b0	core: Prefer JS_AtomToCString() over JS_AtomToString() + JS_ToCString().	2025-06-11 12:46:52 -04:00
Cory McWilliams	b135a210cc	core: Avoid trivial snprintfs.	2025-06-10 21:17:55 -04:00
Cory McWilliams	3a2a829940	ssb: Disabling room support only disables the ability to tunnel through ourselves, not receive tunneled connections.	2025-06-10 20:36:59 -04:00
Cory McWilliams	899605c860	ssb: Actually prevent the sqlite WAL file from growing out of control by truncating when it grows to about 64MB of pages.	2025-06-07 11:39:31 -04:00
Cory McWilliams	92926fa8df	ssb: Chasing intermittent SQLITE_ERROR with multiple processes accessing the database. Switching to sqlite3_prepare_v2 for better errors, but maybe this is also the solution?	2025-05-21 22:05:33 -04:00
Cory McWilliams	8a0e190a86	ssb: Clean up outdated dynamic blob_wants_cache entries.	2025-05-21 17:44:00 -04:00
Cory McWilliams	4de53b9926	core: sqlite checkpoint tweaks. Switch the periodic checkpoint to passive mode. Truncate mode was painful on mobile. I'm seeing it succeed often enough that we don't grow indefinitely. Also fix the print.	2025-05-07 18:15:33 -04:00
Cory McWilliams	d9684c7d62	ssb: Fix culling feeds again.	2025-04-09 22:15:01 -04:00
Cory McWilliams	6f96d4ce65	ssb: More different feed culling logic.	2025-04-09 20:45:23 -04:00
Cory McWilliams	0bcb033349	ssb: This seems like it would explain replication stalls.	2025-03-19 06:46:34 -04:00
Cory McWilliams	c50de0b0f0	ssb: Testing a theory that message sends are piling up.	2025-03-17 12:27:16 -04:00
Cory McWilliams	bb7d2d7ae0	ssb: Update ebt received value on the fly.	2025-03-16 22:04:06 -04:00
Cory McWilliams	862d172ca8	ssb: Pub blobs in the idle queue, too. I'm confused why things aren't moving like I'd expect. This seems more fair.	2025-03-16 19:30:35 -04:00
Cory McWilliams	3660577a23	ssb: I made blob_wants_view too slow. Let's make a real table updated by triggers.	2025-03-12 18:17:11 -04:00
Cory McWilliams	ee33f54745	ssb: Don't send EBT clocks that weren't asked for. Fixes more replication-related warnings.	2025-03-04 12:19:52 -05:00
Cory McWilliams	a74a9fc821	ssb: Clean up some muxrpc-related warnings.	2025-03-02 19:43:50 -05:00
Cory McWilliams	bfec46673d	ssb: Test an experiment that blobs are getting priority over messages during initial sync.	2025-02-22 11:57:57 -05:00
Cory McWilliams	9c9efb845c	ssb: Schedule the clock for reevaluation any time new messages are added. I think this will improve initial replication.	2025-02-21 22:30:14 -05:00
Cory McWilliams	857f47bf55	ssb: Make the invite.use reponse flags match the docs exactly, and add some connection state paranoia.	2025-02-15 20:52:39 -05:00
Cory McWilliams	373d742751	ssb: The docs show the response to invite.use() wrapped in {key: ..., value: ...}, so do that.	2025-02-15 08:19:23 -05:00
Cory McWilliams	575622c522	ssb: Did that from the wrong thread.	2025-02-13 21:37:40 -05:00
Cory McWilliams	6e5b1127f3	ssb: If we don't accept the invite, give up on the connection.	2025-02-12 18:08:40 -05:00
Cory McWilliams	dddec489b9	ssb: Results of staring at unexpected shs disconnections: wire up tunnel.endpoints and blobs.createWants more correctly/thoroughly. And add slightly more context when deliberately disconnected from the remote side.	2025-02-04 21:32:22 -05:00
Cory McWilliams	f824b8988e	ssb: Correctness around tunnel.endpoints.	2025-01-22 18:03:05 -05:00
Cory McWilliams	0f643bfe39	ssb: Complete using an invite, following the pub and posting a pub message. #99	2025-01-20 17:47:30 -05:00
Cory McWilliams	3f3deb665c	ssb: Add a command-line action to generate an invite, and verified that Patchwork can accept it.	2025-01-19 21:00:38 -05:00
Cory McWilliams	97fc22ce57	ssb: Test the one message generated from an invite so far.	2025-01-19 18:38:26 -05:00
Cory McWilliams	616f3ad76d	ssb: Invite support progress. Now the pub accepts the invite, tracks its use, and follows. The client still needs to react.	2025-01-19 17:02:08 -05:00
Cory McWilliams	287c6c06e1	core: More clean shutdown?	2025-01-11 14:48:12 -05:00
Cory McWilliams	02759c6f83	ssb: Hint at follow depth with profile image shape. Also, reload follow information the same way we re-determine channel unread status. Let's see if this feels good.	2025-01-11 13:48:06 -05:00
Cory McWilliams	e59a00922b	ssb: Respond to tunnel.endpoints. Patchwork didn't seem to like that we responded to tunnel.isRoom but not this.	2025-01-11 09:23:12 -05:00
Cory McWilliams	d0bbd7f24f	ssb: Add a has_blob command. #89	2025-01-06 20:46:16 -05:00
Cory McWilliams	aff98110e0	ssb: Tidy up some of the more common reasons for disconnect.	2025-01-04 21:37:43 -05:00
Cory McWilliams	2328f3afb5	ssb: Ease up on excessively re-hitting the database for ebt.replicate even more.	2025-01-04 09:58:16 -05:00
Cory McWilliams	2223245861	ssb: Maybe ease up on hammering the db for follows.	2025-01-03 18:17:54 -05:00
Cory McWilliams	36226b01cd	ssb: Manage new message handling from the new EBT code.	2025-01-03 17:04:38 -05:00
Cory McWilliams	44d9f69434	ssb: Refactoring EBT implementation. I think this works not worse than before and will let me schedule message replication better in a future change. #93	2025-01-03 13:59:25 -05:00
Cory McWilliams	3f343b283b	ssb: Delete one more redundant global setting accessor.	2025-01-03 08:41:13 -05:00
Cory McWilliams	cd2fe9f8d9	ssb: Fix a crash on Windows when we would call uv_async_send on a handle that had already been closed. Various other cleanup and improvements along the journey. #96	2025-01-02 12:35:58 -05:00
Cory McWilliams	2dd32cdce2	ssb: Tweak idle scheduling even more, still. Fixes -t=bench.	2024-12-27 15:51:33 -05:00
Cory McWilliams	9cddd93dad	ssb: Don't schedule duplicate history stream requests for the same account. Changes how we schedule idle work. Let's see if this is better.	2024-12-27 15:02:52 -05:00
Cory McWilliams	4127898655	ssb: Avoid more work on shutdown.	2024-12-27 14:27:52 -05:00
Cory McWilliams	45d48483d0	ssb: Avoid scheduling idle work while shutting down, so that we shut down sooner.	2024-12-27 14:03:14 -05:00
Cory McWilliams	7843168fad	ssb: Hook up some more disconnect messaging.	2024-12-26 20:12:04 -05:00
Cory McWilliams	41e91f2922	ssb: Consolidate global settings helpers.	2024-12-24 11:16:52 -05:00
Cory McWilliams	2f266b8dd4	ssb: Attempt to request more feeds as more contact messages come in. #83	2024-12-11 20:26:28 -05:00