keep replication batches alive for longer (#5931)

* keeping replication batches alive for longer

* fix timeouts, fix revisions

* honor force in gc

arangod/RocksDBEngine/RocksDBCounterManager.cpp

@@ -39,6 +39,7 @@
 #include "RocksDBEngine/RocksDBKeyBounds.h"
 #include "RocksDBEngine/RocksDBVPackIndex.h"
 #include "RocksDBEngine/RocksDBValue.h"
+#include "Transaction/Helpers.h"
 #include "StorageEngine/EngineSelectorFeature.h"
 #include "VocBase/ticks.h"
 
@@ -641,14 +642,12 @@ class WBReader final : public rocksdb::WriteBatch::Handler {
     updateMaxTick(column_family_id, key, value);
     if (shouldHandleDocument(column_family_id, key)) {
       uint64_t objectId = RocksDBKey::objectId(key);
-      uint64_t revisionId =
-          RocksDBKey::revisionId(RocksDBEntryType::Document, key);
 
       auto const& it = deltas.find(objectId);
       if (it != deltas.end()) {
         it->second._sequenceNum = currentSeqNum;
         it->second._added++;
-        it->second._revisionId = revisionId;
+        it->second._revisionId = transaction::helpers::extractRevFromDocument(RocksDBValue::data(value));
       }
     } else {
       // We have to adjust the estimate with an insert
@@ -677,14 +676,11 @@ class WBReader final : public rocksdb::WriteBatch::Handler {
                            const rocksdb::Slice& key) override {
     if (shouldHandleDocument(column_family_id, key)) {
       uint64_t objectId = RocksDBKey::objectId(key);
-      uint64_t revisionId =
-          RocksDBKey::revisionId(RocksDBEntryType::Document, key);
 
       auto const& it = deltas.find(objectId);
       if (it != deltas.end()) {
         it->second._sequenceNum = currentSeqNum;
         it->second._removed++;
-        it->second._revisionId = revisionId;
       }
     } else {
       // We have to adjust the estimate with an remove

arangod/RocksDBEngine/RocksDBReplicationContext.cpp

@@ -525,11 +525,7 @@ void RocksDBReplicationContext::use(double ttl) {
   TRI_ASSERT(!_isUsed);
 
   _isUsed = true;
-  if (_ttl > 0.0) {
+  ttl = std::max(std::max(_ttl, ttl), 300.0);
-    ttl = _ttl;
-  } else {
-    ttl = InitialSyncer::defaultBatchTimeout;
-  }
   _expires = TRI_microtime() + ttl;
   if (_serverId != 0) {
     _vocbase->updateReplicationClient(_serverId, ttl);
@@ -538,6 +534,8 @@ void RocksDBReplicationContext::use(double ttl) {
 
 void RocksDBReplicationContext::release() {
   TRI_ASSERT(_isUsed);
+  double ttl = std::max(_ttl, 300.0);
+  _expires = TRI_microtime() + ttl;
   _isUsed = false;
   if (_serverId != 0) {
     double ttl;

arangod/RocksDBEngine/RocksDBReplicationManager.cpp

@@ -299,18 +299,18 @@ bool RocksDBReplicationManager::garbageCollect(bool force) {
     for (auto it = _contexts.begin(); it != _contexts.end();
          /* no hoisting */) {
       auto context = it->second;
+
+      if (!force && context->isUsed()) {
+        // must not physically destroy contexts that are currently used
+        ++it;
+        continue;
+      }
       
       if (force || context->expires() < now) {
         // expire contexts
         context->deleted();
       }
 
-      if (context->isUsed()) {
-        // must not physically destroy contexts that are currently used
-        ++it;
-        continue;
-      }
-
       if (context->isDeleted()) {
         try {
           found.emplace_back(context);