From 3787ddae894d1962374063a440114237fe8d3abd Mon Sep 17 00:00:00 2001
From: "Bharadwaj V.R" <bharadwaj.vr@snowflake.com>
Date: Fri, 15 Apr 2022 09:04:52 -0700
Subject: [PATCH] Add trace-based status reporting for server version

---
 fdbserver/ClusterRecovery.actor.cpp |  7 ++++++-
 fdbserver/ClusterRecovery.actor.h   |  3 +++
 fdbserver/Status.actor.cpp          | 26 ++++++++++++++++++++++++--
 3 files changed, 33 insertions(+), 3 deletions(-)

diff --git a/fdbserver/ClusterRecovery.actor.cpp b/fdbserver/ClusterRecovery.actor.cpp
index 62696aa9cd..402b82896d 100644
--- a/fdbserver/ClusterRecovery.actor.cpp
+++ b/fdbserver/ClusterRecovery.actor.cpp
@@ -1409,7 +1409,7 @@ ACTOR Future<Void> clusterRecoveryCore(Reference<ClusterRecoveryData> self) {
 	wait(self->cstate.read());
 
 	if (self->cstate.prevDBState.lowestCompatibleServerVersion > currentProtocolVersion) {
-		TraceEvent(SevWarnAlways, "IncompatbleServerVersion", self->dbgid).log();
+		TraceEvent(SevWarnAlways, "IncompatibleServerVersion", self->dbgid).log();
 		throw internal_error();
 	}
 
@@ -1478,6 +1478,11 @@ ACTOR Future<Void> clusterRecoveryCore(Reference<ClusterRecoveryData> self) {
 	}
 	wait(self->cstate.write(newState) || recoverAndEndEpoch);
 
+	TraceEvent("SWVersionCompatibilityChecked", self->dbgid)
+	    .detail("NewestServerVersion", self->cstate.myDBState.newestServerVersion)
+	    .detail("LowestCompatibleVersion", self->cstate.myDBState.lowestCompatibleServerVersion)
+	    .trackLatest(self->swVersionCheckedEventHolder->trackingKey);
+
 	self->recoveryState = RecoveryState::RECRUITING;
 
 	state std::vector<StorageServerInterface> seedServers;
diff --git a/fdbserver/ClusterRecovery.actor.h b/fdbserver/ClusterRecovery.actor.h
index d0deef911f..810bd35f7a 100644
--- a/fdbserver/ClusterRecovery.actor.h
+++ b/fdbserver/ClusterRecovery.actor.h
@@ -22,6 +22,7 @@
 
 // When actually compiled (NO_INTELLISENSE), include the generated version of this file.  In intellisense use the source
 // version.
+#include "flow/Trace.h"
 #include <utility>
 
 #if defined(NO_INTELLISENSE) && !defined(FDBSERVER_CLUSTERRECOVERY_ACTOR_G_H)
@@ -244,6 +245,7 @@ struct ClusterRecoveryData : NonCopyable, ReferenceCounted<ClusterRecoveryData>
 
 	Future<Void> logger;
 
+	Reference<EventCacheHolder> swVersionCheckedEventHolder;
 	Reference<EventCacheHolder> recoveredConfigEventHolder;
 	Reference<EventCacheHolder> clusterRecoveryStateEventHolder;
 	Reference<EventCacheHolder> clusterRecoveryGenerationsEventHolder;
@@ -273,6 +275,7 @@ struct ClusterRecoveryData : NonCopyable, ReferenceCounted<ClusterRecoveryData>
 	    backupWorkerDoneRequests("BackupWorkerDoneRequests", cc),
 	    getLiveCommittedVersionRequests("GetLiveCommittedVersionRequests", cc),
 	    reportLiveCommittedVersionRequests("ReportLiveCommittedVersionRequests", cc),
+	    swVersionCheckedEventHolder(makeReference<EventCacheHolder>("SWVersionCompatibilityChecked")),
 	    recoveredConfigEventHolder(makeReference<EventCacheHolder>("RecoveredConfig")) {
 		clusterRecoveryStateEventHolder = makeReference<EventCacheHolder>(
 		    getRecoveryEventName(ClusterRecoveryEventType::CLUSTER_RECOVERY_STATE_EVENT_NAME));
diff --git a/fdbserver/Status.actor.cpp b/fdbserver/Status.actor.cpp
index ee26a6970d..5fa1600ee7 100644
--- a/fdbserver/Status.actor.cpp
+++ b/fdbserver/Status.actor.cpp
@@ -1543,6 +1543,25 @@ ACTOR Future<ProtocolVersion> getNewestProtocolVersion(Database cx) {
 	}
 }
 
+ACTOR Future<ProtocolVersion> getNewestProtocolVersion(Database cx, WorkerDetails ccWorker) {
+
+	try {
+		state Future<TraceEventFields> swVersionF = timeoutError(
+		    ccWorker.interf.eventLogRequest.getReply(EventLogRequest("SWVersionCompatibilityChecked"_sr)), 1.0);
+
+		wait(success(swVersionF));
+		const TraceEventFields& swVersionTrace = swVersionF.get();
+		int64_t newestProtocolVersionValue = atoi(swVersionTrace.getValue("NewestServerVersion").c_str());
+		return ProtocolVersion(newestProtocolVersionValue);
+
+	} catch (Error& e) {
+		if (e.code() == error_code_actor_cancelled)
+			throw;
+
+		return ProtocolVersion();
+	}
+}
+
 struct LoadConfigurationResult {
 	bool fullReplication;
 	Optional<Key> healthyZone;
@@ -2896,6 +2915,8 @@ ACTOR Future<StatusReply> clusterGetStatus(
 			messages.push_back(message);
 		}
 
+		state ProtocolVersion newestProtocolVersion = wait(getNewestProtocolVersion(cx, ccWorker));
+
 		// construct status information for cluster subsections
 		state int statusCode = (int)RecoveryStatus::END;
 		state JsonBuilderObject recoveryStateStatus = wait(
@@ -2933,10 +2954,11 @@ ACTOR Future<StatusReply> clusterGetStatus(
 		statusObj["protocol_version"] = format("%" PRIx64, g_network->protocolVersion().version());
 		statusObj["connection_string"] = coordinators.ccr->getConnectionString().toString();
 		statusObj["bounce_impact"] = getBounceImpactInfo(statusCode);
-
-		ProtocolVersion newestProtocolVersion = wait(getNewestProtocolVersion(cx));
 		statusObj["latest_server_version"] = format("%" PRIx64, newestProtocolVersion.version());
 
+		// ProtocolVersion newestProtocolVersion = wait(getNewestProtocolVersion(cx));
+		// statusObj["latest_server_version"] = format("%" PRIx64, newestProtocolVersion.version());
+
 		state Optional<DatabaseConfiguration> configuration;
 		state Optional<LoadConfigurationResult> loadResult;