MB-48467: Add config for magma MinCheckpointCreationInterval

Magma will create a checkpoint if the overhead of the checkpoint
queue exceeds 10% of the data size including fragmentation. It is
configured to 2% per checkpoint via magma_checkpoint_threshold
in config.json which results in 10% since magma holds a max of
five checkpoints.

During data load or overwrite workloads where the data size is
small, this can result in agressive checkpoint creation. This
patch adds MinCheckpointCreationInterval config to rate limit
checkpoint creation in such workloads.

Also set the config to zero for tests which require creation of
a checkpoint every batch.

Change-Id: If09b7fd6d7615a583d8f99c050682c66b1842727
Reviewed-on: http://review.couchbase.org/c/kv_engine/+/163750
Reviewed-by: Ben Huddleston <[email protected]>
Tested-by: Build Bot <[email protected]>

Author: Apaar

engines/ep/configuration.json

@@ -1247,7 +1247,13 @@
         "magma_checkpoint_interval": {
             "default": "120",
             "dynamic": false,
-            "descr": "Frequency of checkpoint interval; in seconds. A checkpoint interval provides a rollback point to which the data store can rollback to in the event of a failure.",
+            "descr": "Frequency of checkpoint interval; in seconds. A checkpoint provides a rollback point to which the data store can rollback to in the event of a failure.",
+            "type": "size_t"
+        },
+        "magma_min_checkpoint_interval": {
+            "default": "1",
+            "dynamic": false,
+            "descr": "Minimum interval between two checkpoints; in seconds. Prevents excessive creation of checkpoints.",
             "type": "size_t"
         },
         "magma_checkpoint_threshold": {

engines/ep/src/kvstore/magma-kvstore/magma-kvstore.cc

@@ -494,6 +494,8 @@ MagmaKVStore::MagmaKVStore(MagmaKVStoreConfig& configuration)
             configuration.getMagmaCheckpointInterval();
     configuration.magmaCfg.CheckpointCreationThreshold =
             configuration.getMagmaCheckpointThreshold();
+    configuration.magmaCfg.MinCheckpointCreationInterval =
+            configuration.getMagmaMinCheckpointInterval();
     configuration.magmaCfg.HeartbeatInterval =
             configuration.getMagmaHeartbeatInterval();
     configuration.magmaCfg.MinValueSize =

engines/ep/src/kvstore/magma-kvstore/magma-kvstore_config.cc

@@ -67,6 +67,8 @@ MagmaKVStoreConfig::MagmaKVStoreConfig(Configuration& config,
     magmaMaxCheckpoints = config.getMagmaMaxCheckpoints();
     magmaCheckpointInterval =
             std::chrono::milliseconds(1s * config.getMagmaCheckpointInterval());
+    magmaMinCheckpointInterval = std::chrono::milliseconds(
+            1s * config.getMagmaMinCheckpointInterval());
     magmaCheckpointThreshold = config.getMagmaCheckpointThreshold();
     magmaHeartbeatInterval =
             std::chrono::milliseconds(1s * config.getMagmaHeartbeatInterval());

engines/ep/src/kvstore/magma-kvstore/magma-kvstore_config.h

@@ -57,6 +57,12 @@ class MagmaKVStoreConfig : public KVStoreConfig {
     void setMagmaCheckpointInterval(std::chrono::milliseconds val) {
         magmaCheckpointInterval = val;
     }
+    std::chrono::milliseconds getMagmaMinCheckpointInterval() const {
+        return magmaMinCheckpointInterval;
+    }
+    void setMagmaMinCheckpointInterval(std::chrono::milliseconds val) {
+        magmaMinCheckpointInterval = val;
+    }
     float getMagmaCheckpointThreshold() const {
         return magmaCheckpointThreshold;
     }
@@ -230,6 +236,9 @@ class MagmaKVStoreConfig : public KVStoreConfig {
     // Time interval between checkpoints
     std::chrono::milliseconds magmaCheckpointInterval;
 
+    // Minimum time interval between two checkpoints
+    std::chrono::milliseconds magmaMinCheckpointInterval;
+
     // Fraction of total data before checkpoint is created
     float magmaCheckpointThreshold;
 

engines/ep/src/kvstore/nexus-kvstore/nexus-kvstore-config.cc

@@ -50,6 +50,8 @@ NexusKVStoreConfig::NexusKVStoreConfig(Configuration& config,
         magmaKVStoreConfig.setMagmaSyncEveryBatch(true);
         magmaKVStoreConfig.setMagmaCheckpointInterval(
                 std::chrono::milliseconds(0));
+        magmaKVStoreConfig.setMagmaMinCheckpointInterval(
+                std::chrono::milliseconds(0));
 #endif
     } else if (primaryBackend == "couchdb") {
         // Couchstore has an optimization that forces rollback to zero should
@@ -66,6 +68,8 @@ NexusKVStoreConfig::NexusKVStoreConfig(Configuration& config,
         magmaKVStoreConfig.setMagmaSyncEveryBatch(true);
         magmaKVStoreConfig.setMagmaCheckpointInterval(
                 std::chrono::milliseconds(0));
+        magmaKVStoreConfig.setMagmaMinCheckpointInterval(
+                std::chrono::milliseconds(0));
 #endif
     } else if (primaryBackend == "couchdb") {
         // Couchstore has an optimization that forces rollback to zero should

engines/ep/tests/ep_testsuite.cc

@@ -7011,6 +7011,7 @@ static enum test_result test_mb19687_fixed(EngineIface* h) {
               "ep_item_num_based_new_chk",
               "ep_magma_sync_every_batch",
               "ep_magma_checkpoint_interval",
+              "ep_magma_min_checkpoint_interval",
               "ep_magma_checkpoint_threshold",
               "ep_magma_delete_frag_ratio",
               "ep_magma_delete_memtable_writecache",
@@ -7321,6 +7322,7 @@ static enum test_result test_mb19687_fixed(EngineIface* h) {
               "ep_num_writer_threads",
               "ep_magma_sync_every_batch",
               "ep_magma_checkpoint_interval",
+              "ep_magma_min_checkpoint_interval",
               "ep_magma_checkpoint_threshold",
               "ep_magma_delete_frag_ratio",
               "ep_magma_delete_memtable_writecache",
@@ -8689,6 +8691,7 @@ BaseTestCase testsuite_testcases[] = {
                  test_setup,
                  teardown,
                  "magma_checkpoint_interval=0;"
+                 "magma_min_checkpoint_interval=0;"
                  "magma_sync_every_batch=true",
                  /* TODO RDB: Needs stat:ep_db_data_size */
                  prepare_ep_bucket_skip_broken_under_rocks,
@@ -8698,6 +8701,7 @@ BaseTestCase testsuite_testcases[] = {
                  test_setup,
                  teardown,
                  "magma_checkpoint_interval=0;"
+                 "magma_min_checkpoint_interval=0;"
                  "magma_sync_every_batch=true",
                  prepare,
                  cleanup),
@@ -8779,6 +8783,7 @@ BaseTestCase testsuite_testcases[] = {
                  test_setup,
                  teardown,
                  "magma_checkpoint_interval=0;"
+                 "magma_min_checkpoint_interval=0;"
                  "magma_sync_every_batch=true",
                  /* TODO RDB: DB file size is not reported correctly */
                  prepare_ep_bucket_skip_broken_under_rocks,

engines/ep/tests/ep_testsuite_dcp.cc

@@ -8433,10 +8433,12 @@ BaseTestCase testsuite_testcases[] = {
                  test_chk_manager_rollback,
                  test_setup,
                  teardown,
-                 // 'magma_checkpoint_interval=0' allows us to create more than
-                 // one checkpoint in less than 2mins
+                 // 'magma_checkpoint_interval=0' and
+                 // 'magma_min_checkpoint_interval=0' allows us to create more
+                 // than one checkpoint in less than 2mins
                  "dcp_flow_control_policy=none;dcp_enable_noop=false;"
-                 "magma_checkpoint_interval=0;",
+                 "magma_checkpoint_interval=0;"
+                 "magma_min_checkpoint_interval=0;",
                  // TODO RDB: implement getItemCount.
                  // Needs the 'curr_items_tot' stat.
                  prepare_skip_broken_under_rocks,
@@ -8456,13 +8458,14 @@ BaseTestCase testsuite_testcases[] = {
                 test_partialrollback_for_consumer,
                 test_setup,
                 teardown,
-                // 'magma_checkpoint_interval=0' allows us to create more than
-                // one checkpoint in less than 2mins.
-                // 'magma_max_checkpoints=10' the max number of checkpoints that
-                // can be rolled back.
+                // 'magma_checkpoint_interval=0;magma_min_checkpoint_interval=0;'
+                // allows us to create more than one checkpoint in less than
+                // 2mins. 'magma_max_checkpoints=10' the max number of
+                // checkpoints that can be rolled back.
                 // 'magma_sync_every_batch=true' makes magma behaviour
                 // like couchstore, creating a checkpoint for every flush batch.
-                "dcp_enable_noop=false;magma_checkpoint_interval=0;"
+                "dcp_enable_noop=false;"
+                "magma_checkpoint_interval=0;magma_min_checkpoint_interval=0;"
                 "magma_max_checkpoints=10;magma_sync_every_batch=true",
                 // TODO RDB: implement getItemCount.
                 // Needs the 'vb_replica_curr_items' stat.

engines/ep/tests/module_tests/item_pager_test.cc

@@ -59,6 +59,7 @@ class STBucketQuotaTest : public STParameterizedBucketTest {
         config_string +=
                 "ht_size=47;"
                 "magma_checkpoint_interval=0;"
+                "magma_min_checkpoint_interval=0;"
                 "magma_sync_every_batch=true";
 
         STParameterizedBucketTest::SetUp();

engines/ep/tests/module_tests/test_helpers.h

@@ -39,6 +39,7 @@ static std::string magmaConfig =
 static std::string magmaRollbackConfig =
         "magma_max_checkpoints=10;"
         "magma_checkpoint_interval=0;"
+        "magma_min_checkpoint_interval=0;"
         "magma_sync_every_batch=true";
 
 class VBucket;