Commit 059d1e1
Fixed RPC subscriptions leak when subscription stream is finished (paritytech#4533)
closes paritytech/parity-bridges-common#3000
Recently we've changed our bridge configuration for Rococo <> Westend
and our new relayer has started to submit transactions every ~ `30`
seconds. Eventually, it switches itself into limbo state, where it can't
submit more transactions - all `author_submitAndWatchExtrinsic` calls
are failing with the following error: `ERROR bridge Failed to send
transaction to BridgeHubRococo node: Call(ErrorObject { code:
ServerError(-32006), message: "Too many subscriptions on the
connection", data: Some(RawValue("Exceeded max limit of 1024")) })`.
Some links for those who want to explore:
- server side (node) has a strict limit on a number of active
subscriptions. It fails to open a new subscription if this limit is hit:
https://github.com/paritytech/jsonrpsee/blob/a4533966b997e83632509ad97eea010fc7c3efc0/server/src/middleware/rpc/layer/rpc_service.rs#L122-L132.
The limit is set to `1024` by default;
- internally this limit is a semaphore with `limit` permits:
https://github.com/paritytech/jsonrpsee/blob/a4533966b997e83632509ad97eea010fc7c3efc0/core/src/server/subscription.rs#L461-L485;
- semaphore permit is acquired in the first link;
- the permit is "returned" when the `SubscriptionSink` is dropped:
https://github.com/paritytech/jsonrpsee/blob/a4533966b997e83632509ad97eea010fc7c3efc0/core/src/server/subscription.rs#L310-L325;
- the `SubscriptionSink` is dropped when [this `polkadot-sdk`
function](https://github.com/paritytech/polkadot-sdk/blob/278486f9bf7db06c174203f098eec2f91839757a/substrate/client/rpc/src/utils.rs#L58-L94)
returns. In other words - when the connection is closed, the stream is
finished or internal subscription buffer limit is hit;
- the subscription has the internal buffer, so sending an item contains
of two steps: [reading an item from the underlying
stream](https://github.com/paritytech/polkadot-sdk/blob/278486f9bf7db06c174203f098eec2f91839757a/substrate/client/rpc/src/utils.rs#L125-L141)
and [sending it over the
connection](https://github.com/paritytech/polkadot-sdk/blob/278486f9bf7db06c174203f098eec2f91839757a/substrate/client/rpc/src/utils.rs#L111-L116);
- when the underlying stream is finished, the `inner_pipe_from_stream`
wants to ensure that all items are sent to the subscriber. So it: [waits
until the current send operation
completes](https://github.com/paritytech/polkadot-sdk/blob/278486f9bf7db06c174203f098eec2f91839757a/substrate/client/rpc/src/utils.rs#L146-L148)
and then [send all remaining items from the internal
buffer](https://github.com/paritytech/polkadot-sdk/blob/278486f9bf7db06c174203f098eec2f91839757a/substrate/client/rpc/src/utils.rs#L150-L155).
Once it is done, the function returns, the `SubscriptionSink` is
dropped, semaphore permit is dropped and we are ready to accept new
subscriptions;
- unfortunately, the code just calls the `pending_fut.await.is_err()` to
ensure that [the current send operation
completes](https://github.com/paritytech/polkadot-sdk/blob/278486f9bf7db06c174203f098eec2f91839757a/substrate/client/rpc/src/utils.rs#L146-L148).
But if there are no current send operation (which is normal), then the
`pending_fut` is set to terminated future and the `await` never
completes. Hence, no return from the function, no drop of
`SubscriptionSink`, no drop of semaphore permit, no new subscriptions
allowed (once number of susbcriptions hits the limit.
I've illustrated the issue with small test - you may ensure that if e.g.
the stream is initially empty, the
`subscription_is_dropped_when_stream_is_empty` will hang because
`pipe_from_stream` never exits.1 parent 3d24d5a commit 059d1e1
2 files changed
Lines changed: 35 additions & 1 deletion
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
| 1 | + | |
| 2 | + | |
| 3 | + | |
| 4 | + | |
| 5 | + | |
| 6 | + | |
| 7 | + | |
| 8 | + | |
| 9 | + | |
| 10 | + | |
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
143 | 143 | | |
144 | 144 | | |
145 | 145 | | |
146 | | - | |
| 146 | + | |
147 | 147 | | |
148 | 148 | | |
149 | 149 | | |
| |||
231 | 231 | | |
232 | 232 | | |
233 | 233 | | |
| 234 | + | |
| 235 | + | |
| 236 | + | |
| 237 | + | |
| 238 | + | |
| 239 | + | |
| 240 | + | |
| 241 | + | |
| 242 | + | |
| 243 | + | |
| 244 | + | |
| 245 | + | |
| 246 | + | |
| 247 | + | |
| 248 | + | |
| 249 | + | |
| 250 | + | |
| 251 | + | |
| 252 | + | |
| 253 | + | |
| 254 | + | |
| 255 | + | |
| 256 | + | |
| 257 | + | |
234 | 258 | | |
0 commit comments