fix(network): Verify edge on separate thread #3145
Conversation
|
|
There was a problem hiding this comment.
@mfornet is this enough to address the issue or do we also need to fix the schedule that recomputes the graph?
| @@ -67,7 +67,7 @@ async def main(): | |||
| assert response.Handshake.target_peer_id.data == bytes( | |||
| my_key_pair_nacl.verify_key) | |||
| assert response.Handshake.listen_port == nodes[0].addr()[1] | |||
| assert response.Handshake.version == handshake.Handshake.version | |||
| # assert response.Handshake.version == handshake.Handshake.version | |||
There was a problem hiding this comment.
Actually that line shouldn't be removed, but it will fail until #3016 is merged. Will add a todo
| @@ -0,0 +1,115 @@ | |||
| """ | |||
There was a problem hiding this comment.
Can we add this test to nightly?
There was a problem hiding this comment.
Not yet, since it doesn't have a failure/successful condition.
There was a problem hiding this comment.
I was using this test for debugging, the condition will be probably based on logs from delay-detector, will add in next PR, since we need too push this ASAP
I'm also fixing the schedule that recompute the graph, but this PR fixes our current issue. Routing table update is very cheap even with ~6K edges, it takes less than 100ms. |
|
@mfornet looks like it doesn't compile https://buildkite.com/nearprotocol/nearcore/builds/3027#71fc28f4-5299-4298-a4e3-be056abdc5f1 |
mfornet
force-pushed
the
routing_table_threaded
branch
from
4045273 to
33b2269
Compare
Codecov Report
@@ Coverage Diff @@
## master #3145 +/- ##
=======================================
Coverage 87.34% 87.34%
=======================================
Files 212 212
Lines 41525 41525
=======================================
Hits 36270 36270
Misses 5255 5255 Continue to review full report at Codecov.
|
* fix(network): Verify edge on separate thread * Address comments from bowen * Fix compile error
* fix(network): Verify edge on separate thread * Address comments from bowen * Fix compile error
Sync message with large number of edges was blocking the PeerManagerActor. This was because verifying signature for each edge is a very expensive computation, for a message with 2K edges it takes ~6seconds. This computation was moved into a separate thread so the PeerManagerActor is not blocked.
TestPlan
Compile
neardwithdelay_detector_enabledand runsaturate_routing_table.py. Check that no messages block the PeerManagerActor for more than 50ms.