p2p: fan in incoming txns into backlog worker #6126
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #6126 +/- ##
==========================================
- Coverage 51.73% 50.77% -0.97%
==========================================
Files 644 644
Lines 86519 86529 +10
==========================================
- Hits 44763 43937 -826
- Misses 38889 39683 +794
- Partials 2867 2909 +42 ☔ View full report in Codecov by Sentry. |
algorandskiy
force-pushed
the
pavel/p2p-txhander-bg
branch
4 times, most recently
from
a65acca to
2e055ca
Compare
algorandskiy
changed the title
algorandskiy
force-pushed
the
pavel/p2p-txhander-bg
branch
from
5759b58 to
5635798
Compare
|
Merged master in. |
| @@ -59,9 +61,18 @@ type SignedTxnWithAD struct { | |||
|
|
|||
| // ID returns the Txid (i.e., hash) of the underlying transaction. | |||
| func (s SignedTxn) ID() Txid { | |||
| if s.cachedID != nil { | |||
There was a problem hiding this comment.
Curious: did we test performance both with and without the cachedID change?
There was a problem hiding this comment.
I do not recall but can perf tests
| } | ||
|
|
||
| if handler.checkAlreadyCommitted(wi) { |
There was a problem hiding this comment.
So we are cutting down on duplicated logic between what the backlogWorker is doing vs incoming message validation.
|
|
||
| transactionMessagesRemember.Inc(nil) |
There was a problem hiding this comment.
relying on backlog worker for this?
| @@ -2774,6 +2774,8 @@ func TestTxHandlerValidateIncomingTxMessage(t *testing.T) { | |||
|
|
|||
| handler, err := makeTestTxHandler(ledger, cfg) | |||
| require.NoError(t, err) | |||
| handler.Start() | |||
| defer handler.Stop() | |||
There was a problem hiding this comment.
The handler variable gets set to a new test handler on line 2810 - which handler will this refer to when stop() is called (I would think latest referenced object).
| if err != nil { | ||
| if err != pubsub.ErrSubscriptionCancelled && err != context.Canceled { | ||
| n.log.Errorf("Error reading from subscription %v, peerId %s", err, n.service.ID()) | ||
| const threads = incomingThreads / 2 // perf tests showed that 10 (half of incomingThreads) was optimal in terms of TPS (attempted 1, 5, 10, 20) |
There was a problem hiding this comment.
This was on our default specced hardware?
Summary
While investigating p2p TX traffic and performance, I found transaction pool mutex congestion. This PR is a PoC to use
backlogWorkeras a pool only accessor similarly to wsnet.Implementation summary:
syncChchannel to work itemwisyncChis initiated and awaited making validation synchronous as before.backlogWorkerchecks ifsyncChis set and responds with validation result to the channel.Additionally, there are couple more fixes that helped with TPS as well:
sub.Next()faster.Test Plan