This repository was archived by the owner on Jan 14, 2022. It is now read-only.
Tags: mastodon/flodgatt
Tags
Add tests for polling for multiple messages (#149) * Add tests for polling for multiple messages This commit adds a mock Redis interface and adds tests that poll the mock interface for multiple messages at a time. These tests test that Flodgatt is robust against receiving incomplete messages, including if the message break results in receiving invalid UTF8. * Remove temporary files
Improve handling of large Redis input (#143) * Implement faster buffered input This commit implements a modified ring buffer for input from Redis. Specifically, Flodgatt now limits the amount of data it fetches from Redis in one syscall to 8 KiB (two pages on most systems). Flodgatt will process all complete messages it receives from Redis and then re-use the same buffer for the next time it retrieves data. If Flodgatt received a partial message, it will copy the partial message to the beginning of the buffer before its next read. This change has little effect on Flodgatt under light load (because it was rare for Redis to have more than 8 KiB of messages available at any one time). However, my hope is that this will significantly reduce memory use on the largest instances. * Improve handling of backpresure This commit alters how Flodgatt behaves if it receives enough messages for a single client to fill that clients channel. (Because the clients regularly send their messages, should only occur if a single client receives a large number of messages nearly simultaneously; this is rare, but could occur, especially on large instances). Previously, Flodgatt would drop messages in the rare case when the client's channel was full. Now, Flodgatt will pause the current Redis poll and yield control back to the client streams, allowing the clients to empty their channels; Flodgatt will then resume polling Redis/sending the messages it previously received. With the approach, Flodgatt will never drop messages. However, the risk to this approach is that, by never dropping messages, Flodgatt does not have any way to reduce the amount of work it needs to do when under heavy load – it delays the work slightly, but doesn't reduce it. What this means is that it would be *theoretically* possible for Flodgatt to fall increasingly behind, if it is continuously receiving more messages than it can process. Due to how quickly Flodgatt can process messages, though, I suspect this would only come up if an admin were running Flodgatt in a *significantly* resource constrained environment, but I wanted to mention it for the sake of completeness. This commit also adds a new /status/backpressure endpoint that displays the current length of the Redis input buffer (which should typically be low or 0). Like the other /status endpoints, this endpoint is only enabled when Flodgatt is compiled with the `stub_status` feature.
Improve handling of large Redis input (#143) * Implement faster buffered input This commit implements a modified ring buffer for input from Redis. Specifically, Flodgatt now limits the amount of data it fetches from Redis in one syscall to 8 KiB (two pages on most systems). Flodgatt will process all complete messages it receives from Redis and then re-use the same buffer for the next time it retrieves data. If Flodgatt received a partial message, it will copy the partial message to the beginning of the buffer before its next read. This change has little effect on Flodgatt under light load (because it was rare for Redis to have more than 8 KiB of messages available at any one time). However, my hope is that this will significantly reduce memory use on the largest instances. * Improve handling of backpresure This commit alters how Flodgatt behaves if it receives enough messages for a single client to fill that clients channel. (Because the clients regularly send their messages, should only occur if a single client receives a large number of messages nearly simultaneously; this is rare, but could occur, especially on large instances). Previously, Flodgatt would drop messages in the rare case when the client's channel was full. Now, Flodgatt will pause the current Redis poll and yield control back to the client streams, allowing the clients to empty their channels; Flodgatt will then resume polling Redis/sending the messages it previously received. With the approach, Flodgatt will never drop messages. However, the risk to this approach is that, by never dropping messages, Flodgatt does not have any way to reduce the amount of work it needs to do when under heavy load – it delays the work slightly, but doesn't reduce it. What this means is that it would be *theoretically* possible for Flodgatt to fall increasingly behind, if it is continuously receiving more messages than it can process. Due to how quickly Flodgatt can process messages, though, I suspect this would only come up if an admin were running Flodgatt in a *significantly* resource constrained environment, but I wanted to mention it for the sake of completeness. This commit also adds a new /status/backpressure endpoint that displays the current length of the Redis input buffer (which should typically be low or 0). Like the other /status endpoints, this endpoint is only enabled when Flodgatt is compiled with the `stub_status` feature.
Stream events via a watch channel (#128) This squashed commit makes a fairly significant structural change to significantly reduce Flodgatt's CPU usage. Flodgatt connects to Redis in a single (green) thread, and then creates a new thread to handle each WebSocket/SSE connection. Previously, each thread was responsible for polling the Redis thread to determine whether it had a message relevant to the connected client. I initially selected this structure both because it was simple and because it minimized memory overhead – no messages are sent to a particular thread unless they are relevant to the client connected to the thread. However, I recently ran some load tests that show this approach to have unacceptable CPU costs when 300+ clients are simultaneously connected. Accordingly, Flodgatt now uses a different structure: the main Redis thread now announces each incoming message via a watch channel connected to every client thread, and each client thread filters out irrelevant messages. In theory, this could lead to slightly higher memory use, but tests I have run so far have not found a measurable increase. On the other hand, Flodgatt's CPU use is now an order of magnitude lower in tests I've run. This approach does run a (very slight) risk of dropping messages under extremely heavy load: because a watch channel only stores the most recent message transmitted, if Flodgatt adds a second message before the thread can read the first message, the first message will be overwritten and never transmitted. This seems unlikely to happen in practice, and we can avoid the issue entirely by changing to a broadcast channel when we upgrade to the most recent Tokio version (see #75).
Extract tests to separate files (#113) This very minor change moves tests from their current location in submodules within the file under test into submodules in separate files. This is a slight deviation from the normal Rust convention (though only very slight, since the module structure remains the same). However, it is justified here since the tests are fairly verbose and including them in the same file was a bit unwieldy.
PreviousNext