From 85a02f75b169303373cf38bd7134fcb14544135a Mon Sep 17 00:00:00 2001
From: chris erway <51567+cce@users.noreply.github.com>
Date: Tue, 12 Apr 2022 16:02:08 -0400
Subject: [PATCH] add counters around dropping pseudonode and voteVerifier
 tasks, and slow pseudonode responses (#3861)

---
 agreement/cryptoVerifier.go | 5 +++++
 agreement/pseudonode.go     | 9 +++++++++
 2 files changed, 14 insertions(+)

diff --git a/agreement/cryptoVerifier.go b/agreement/cryptoVerifier.go
index ff8a6d6aa4..cf6c466e5d 100644
--- a/agreement/cryptoVerifier.go
+++ b/agreement/cryptoVerifier.go
@@ -22,8 +22,12 @@ import (
 
 	"github.com/algorand/go-algorand/logging"
 	"github.com/algorand/go-algorand/protocol"
+	"github.com/algorand/go-algorand/util/metrics"
 )
 
+var voteVerifierOutFullCounter = metrics.MakeCounter(
+	metrics.MetricName{Name: "algod_agreement_vote_verifier_responses_dropped", Description: "Number of voteVerifier responses dropped due to full channel"})
+
 // TODO put these in config
 const (
 	voteParallelism     = 16
@@ -210,6 +214,7 @@ func (c *poolCryptoVerifier) voteFillWorker(toBundleWait chan<- bundleFuture) {
 				select {
 				case c.votes.out <- asyncVerifyVoteResponse{index: votereq.TaskIndex, err: err, cancelled: true}:
 				default:
+					voteVerifierOutFullCounter.Inc(nil)
 					c.log.Infof("poolCryptoVerifier.voteFillWorker unable to write failed enqueue response to output channel")
 				}
 			}
diff --git a/agreement/pseudonode.go b/agreement/pseudonode.go
index f52854d6a5..bdaa2f359d 100644
--- a/agreement/pseudonode.go
+++ b/agreement/pseudonode.go
@@ -29,6 +29,7 @@ import (
 	"github.com/algorand/go-algorand/logging/logspec"
 	"github.com/algorand/go-algorand/logging/telemetryspec"
 	"github.com/algorand/go-algorand/protocol"
+	"github.com/algorand/go-algorand/util/metrics"
 )
 
 // TODO put these in config
@@ -43,6 +44,9 @@ var errPseudonodeVerifierClosedChannel = errors.New("crypto verifier closed the
 var errPseudonodeNoVotes = errors.New("no valid participation keys to generate votes for given round")
 var errPseudonodeNoProposals = errors.New("no valid participation keys to generate proposals for given round")
 
+var pseudonodeBacklogFullByType = metrics.NewTagCounter("algod_agreement_pseudonode_tasks_dropped_{TAG}", "Number of pseudonode tasks dropped per type")
+var pseudonodeResultTimeoutsByType = metrics.NewTagCounter("algod_agreement_pseudonode_tasks_timeouts_{TAG}", "Number of pseudonode task result timeouts per type")
+
 // A pseudonode creates proposals and votes with a KeyManager which holds participation keys.
 //
 // It constructs these messages as if they arrived from an external source and were verified.
@@ -176,6 +180,7 @@ func (n asyncPseudonode) MakeProposals(ctx context.Context, r round, p period) (
 		return proposalTask.outputChannel(), nil
 	default:
 		proposalTask.close()
+		pseudonodeBacklogFullByType.Add("proposal", 1)
 		return nil, fmt.Errorf("unable to make proposal for (%d, %d): %w", r, p, errPseudonodeBacklogFull)
 	}
 }
@@ -193,6 +198,7 @@ func (n asyncPseudonode) MakeVotes(ctx context.Context, r round, p period, s ste
 		return proposalTask.outputChannel(), nil
 	default:
 		proposalTask.close()
+		pseudonodeBacklogFullByType.Add("vote", 1)
 		return nil, fmt.Errorf("unable to make vote for (%d, %d, %d): %w", r, p, s, errPseudonodeBacklogFull)
 	}
 }
@@ -474,6 +480,7 @@ verifiedVotesLoop:
 				return
 			case <-outputTimeout:
 				// we've been waiting for too long for this vote to be written to the output.
+				pseudonodeResultTimeoutsByType.Add("vote", 1)
 				t.node.log.Warnf("pseudonode.makeVotes: unable to write vote to output channel for round %d, period %d", t.round, t.period)
 				outputTimeout = nil
 			}
@@ -577,6 +584,7 @@ verifiedVotesLoop:
 				return
 			case <-outputTimeout:
 				// we've been waiting for too long for this vote to be written to the output.
+				pseudonodeResultTimeoutsByType.Add("pvote", 1)
 				t.node.log.Warnf("pseudonode.makeProposals: unable to write proposal vote to output channel for round %d, period %d", t.round, t.period)
 				outputTimeout = nil
 			}
@@ -597,6 +605,7 @@ verifiedPayloadsLoop:
 				return
 			case <-outputTimeout:
 				// we've been waiting for too long for this vote to be written to the output.
+				pseudonodeResultTimeoutsByType.Add("ppayload", 1)
 				t.node.log.Warnf("pseudonode.makeProposals: unable to write proposal payload to output channel for round %d, period %d", t.round, t.period)
 				outputTimeout = nil
 			}