SERVER-17003 Add WriteConflict retry logic to applyOps command

1 files changed, 42 insertions, 8 deletions

diff --git a/src/mongo/db/commands/apply_ops.cpp b/src/mongo/db/commands/apply_ops.cpp
index d2522ba6c29..2cc5794c61b 100644
--- a/src/mongo/db/commands/apply_ops.cpp
+++ b/src/mongo/db/commands/apply_ops.cpp
@@ -26,6 +26,10 @@
 *    it in the license file.
 */
 
+#define MONGO_LOG_DEFAULT_COMPONENT ::mongo::logger::LogComponent::kCommand
+
+#include "mongo/platform/basic.h"
+
 #include <sstream>
 #include <string>
 #include <vector>
@@ -38,11 +42,13 @@
 #include "mongo/db/client.h"
 #include "mongo/db/commands.h"
 #include "mongo/db/commands/dbhash.h"
+#include "mongo/db/concurrency/write_conflict_exception.h"
 #include "mongo/db/dbdirectclient.h"
 #include "mongo/db/jsobj.h"
 #include "mongo/db/matcher/matcher.h"
 #include "mongo/db/operation_context_impl.h"
 #include "mongo/db/repl/oplog.h"
+#include "mongo/util/log.h"
 
 namespace mongo {
 
@@ -149,11 +155,26 @@ namespace mongo {
                 };
 
                 Client::Context ctx(txn, ns);
-                bool failed = repl::applyOperation_inlock(txn,
-                                                          ctx.db(),
-                                                          temp,
-                                                          false,
-                                                          alwaysUpsert);
+
+                bool failed;
+                while (true) {
+                    try {
+                        // We assume that in the WriteConflict retry case, either the op rolls back
+                        // any changes it makes or is otherwise safe to rerun.
+                        failed = repl::applyOperation_inlock(txn,
+                                                             ctx.db(),
+                                                             temp,
+                                                             false,
+                                                             alwaysUpsert);
+                        break;
+                    }
+                    catch (const WriteConflictException& wce) {
+                        LOG(2) << "WriteConflictException in applyOps command, retrying.";
+                        txn->recoveryUnit()->commitAndRestart();
+                        continue;
+                    }
+                }
+
                 ab.append(!failed);
                 if ( failed )
                     errors++;
@@ -184,13 +205,26 @@ namespace mongo {
                     }
                 }
 
+                const BSONObj cmdRewritten = cmdBuilder.done();
+
                 // We currently always logOp the command regardless of whether the individial ops
                 // succeeded and rely on any failures to also happen on secondaries. This isn't
                 // perfect, but it's what the command has always done and is part of its "correct"
                 // behavior.
-                WriteUnitOfWork wunit(txn);
-                repl::logOp(txn, "c", tempNS.c_str(), cmdBuilder.done());
-                wunit.commit();
+                while (true) {
+                    try {
+                        WriteUnitOfWork wunit(txn);
+                        repl::logOp(txn, "c", tempNS.c_str(), cmdRewritten);
+                        wunit.commit();
+                        break;
+                    }
+                    catch (const WriteConflictException& wce) {
+                        LOG(2) <<
+                            "WriteConflictException while logging applyOps command, retrying.";
+                        txn->recoveryUnit()->commitAndRestart();
+                        continue;
+                    }
+                }
             }
 
             if (errors != 0) {