SERVER-45953 Exempt internal replication oplog readers from acquiring read tickets in order to avoid deadlocks

(cherry picked from commit d1cb83d9199c1a25158d74e47b3aa88b5c33fe8b)

Conflicts:
        src/mongo/db/commands/getmore_cmd.cpp

2 files changed, 24 insertions, 2 deletions

diff --git a/src/mongo/db/commands/find_cmd.cpp b/src/mongo/db/commands/find_cmd.cpp
index e7313d834f9..50384636a63 100644
--- a/src/mongo/db/commands/find_cmd.cpp
+++ b/src/mongo/db/commands/find_cmd.cpp
@@ -316,11 +316,22 @@ public:
                     !qr->getReadAtClusterTime() || storageEngine->supportsDocLocking());
 
             // Validate term before acquiring locks, if provided.
-            if (auto term = qr->getReplicationTerm()) {
+            auto term = qr->getReplicationTerm();
+            if (term) {
                 // Note: updateTerm returns ok if term stayed the same.
                 uassertStatusOK(replCoord->updateTerm(opCtx, *term));
             }
 
+            // The presence of a term in the request indicates that this is an internal replication
+            // oplog read request.
+            if (term && parsedNss == NamespaceString::kRsOplogNamespace) {
+                // We do not want to take tickets for internal (replication) oplog reads. Stalling
+                // on ticket acquisition can cause complicated deadlocks. Primaries may depend on
+                // data reaching secondaries in order to proceed; and secondaries may get stalled
+                // replicating because of an inability to acquire a read ticket.
+                opCtx->lockState()->skipAcquireTicket();
+            }
+
             // We call RecoveryUnit::setTimestampReadSource() before acquiring a lock on the
             // collection via AutoGetCollectionForRead in order to ensure the comparison to the
             // collection's minimum visible snapshot is accurate.
diff --git a/src/mongo/db/commands/getmore_cmd.cpp b/src/mongo/db/commands/getmore_cmd.cpp
index af923252d7a..450fc19c4bd 100644
--- a/src/mongo/db/commands/getmore_cmd.cpp
+++ b/src/mongo/db/commands/getmore_cmd.cpp
@@ -622,11 +622,22 @@ public:
             auto curOp = CurOp::get(opCtx);
             curOp->debug().cursorid = _request.cursorid;
 
-            // Validate term before acquiring locks, if provided.
             if (_request.term) {
+                // Validate term before acquiring locks.
                 auto replCoord = repl::ReplicationCoordinator::get(opCtx);
                 // Note: updateTerm returns ok if term stayed the same.
                 uassertStatusOK(replCoord->updateTerm(opCtx, *_request.term));
+
+                // If this is an oplog request, then this is a getMore for replication oplog
+                // fetching. The term field is only allowed for internal clients (see
+                // checkAuthForGetMore).
+                if (_request.nss == NamespaceString::kRsOplogNamespace) {
+                    // We do not want to take tickets for internal (replication) oplog reads.
+                    // Stalling on ticket acquisition can cause complicated deadlocks. Primaries may
+                    // depend on data reaching secondaries in order to proceed; and secondaries may
+                    // get stalled replicating because of an inability to acquire a read ticket.
+                    opCtx->lockState()->skipAcquireTicket();
+                }
             }
 
             auto cursorManager = CursorManager::get(opCtx);