From 70322d1ff805f2c8a1a400b4edc3cffb06a8a4ad Mon Sep 17 00:00:00 2001 From: carllin Date: Sat, 16 Nov 2019 02:54:51 -0800 Subject: [PATCH] Add error logging to dead slots (#7000) --- core/src/replay_stage.rs | 6 +++++- .../grafana-provisioning/dashboards/testnet-monitor.json | 2 +- 2 files changed, 6 insertions(+), 2 deletions(-) diff --git a/core/src/replay_stage.rs b/core/src/replay_stage.rs index 0151b1e30..2dd73462b 100644 --- a/core/src/replay_stage.rs +++ b/core/src/replay_stage.rs @@ -488,7 +488,11 @@ impl ReplayStage { bank.slot(), replay_result ); - datapoint_warn!("replay-stage-mark_dead_slot", ("slot", bank.slot(), i64),); + datapoint_error!( + "replay-stage-mark_dead_slot", + ("error", format!("error: {:?}", replay_result), String), + ("slot", bank.slot(), i64) + ); Self::mark_dead_slot(bank.slot(), blocktree, progress); } diff --git a/metrics/scripts/grafana-provisioning/dashboards/testnet-monitor.json b/metrics/scripts/grafana-provisioning/dashboards/testnet-monitor.json index 873f7945b..b85987bfe 100644 --- a/metrics/scripts/grafana-provisioning/dashboards/testnet-monitor.json +++ b/metrics/scripts/grafana-provisioning/dashboards/testnet-monitor.json @@ -4109,7 +4109,7 @@ "hide": false, "orderByTime": "ASC", "policy": "default", - "query": "SELECT host_id, slot FROM \"$testnet\".\"autogen\".\"replay-stage-mark_dead_slot\" WHERE $timeFilter ORDER BY time DESC ", + "query": "SELECT host_id, error, slot FROM \"$testnet\".\"autogen\".\"replay-stage-mark_dead_slot\" WHERE $timeFilter ORDER BY time DESC ", "rawQuery": true, "refId": "A", "resultFormat": "table",