Try to make parallel replicas sleepy

ClickHouse · Apr 30, 2024 · 02f1cce · 02f1cce
1 parent 71589c2
commit 02f1cce
Show file tree

Hide file tree

Showing 4 changed files with 28 additions and 1 deletion.
diff --git a/docker/test/stateful/run.sh b/docker/test/stateful/run.sh
@@ -76,6 +76,26 @@ function start()
         --keeper_server.tcp_port 29181 --keeper_server.server_id 3
     fi
 
+    if [[ -n "$USE_PARALLEL_REPLICAS" ]] && [[ "$USE_PARALLEL_REPLICAS" -eq 1 ]]; then
+        export THREAD_FUZZER_CPU_TIME_PERIOD_US=1000
+        export THREAD_FUZZER_SLEEP_PROBABILITY=0.1
+        export THREAD_FUZZER_SLEEP_TIME_US_MAX=100000
+
+        export THREAD_FUZZER_pthread_mutex_lock_BEFORE_MIGRATE_PROBABILITY=1
+        export THREAD_FUZZER_pthread_mutex_lock_AFTER_MIGRATE_PROBABILITY=1
+        export THREAD_FUZZER_pthread_mutex_unlock_BEFORE_MIGRATE_PROBABILITY=1
+        export THREAD_FUZZER_pthread_mutex_unlock_AFTER_MIGRATE_PROBABILITY=1
+
+        export THREAD_FUZZER_pthread_mutex_lock_BEFORE_SLEEP_PROBABILITY=0.001
+        export THREAD_FUZZER_pthread_mutex_lock_AFTER_SLEEP_PROBABILITY=0.001
+        export THREAD_FUZZER_pthread_mutex_unlock_BEFORE_SLEEP_PROBABILITY=0.001
+        export THREAD_FUZZER_pthread_mutex_unlock_AFTER_SLEEP_PROBABILITY=0.001
+        export THREAD_FUZZER_pthread_mutex_lock_BEFORE_SLEEP_TIME_US_MAX=10000
+        export THREAD_FUZZER_pthread_mutex_lock_AFTER_SLEEP_TIME_US_MAX=10000
+        export THREAD_FUZZER_pthread_mutex_unlock_BEFORE_SLEEP_TIME_US_MAX=10000
+        export THREAD_FUZZER_pthread_mutex_unlock_AFTER_SLEEP_TIME_US_MAX=10000
+    fi
+
     counter=0
     until clickhouse-client --query "SELECT 1"
     do

diff --git a/src/Storages/MergeTree/MergeTreeReadPoolParallelReplicas.cpp b/src/Storages/MergeTree/MergeTreeReadPoolParallelReplicas.cpp
@@ -1,6 +1,6 @@
 #include <iterator>
 #include <Storages/MergeTree/MergeTreeReadPoolParallelReplicas.h>
-
+#include <Common/ThreadFuzzer.h>
 
 namespace DB
 {
@@ -45,6 +45,7 @@ MergeTreeReadTaskPtr MergeTreeReadPoolParallelReplicas::getTask(size_t /*task_id
     if (no_more_tasks_available)
         return nullptr;
 
+    ThreadFuzzer::maybeInjectSleep();
     if (buffered_ranges.empty())
     {
         auto result = extension.callback(ParallelReadRequest(

diff --git a/src/Storages/MergeTree/MergeTreeReadPoolParallelReplicasInOrder.cpp b/src/Storages/MergeTree/MergeTreeReadPoolParallelReplicasInOrder.cpp
@@ -1,4 +1,5 @@
 #include <Storages/MergeTree/MergeTreeReadPoolParallelReplicasInOrder.h>
+#include <Common/ThreadFuzzer.h>
 
 namespace DB
 {
@@ -70,6 +71,7 @@ MergeTreeReadTaskPtr MergeTreeReadPoolParallelReplicasInOrder::getTask(size_t ta
         return std::nullopt;
     };
 
+    ThreadFuzzer::maybeInjectSleep();
     if (auto result = get_from_buffer(); result)
         return createTask(per_part_infos[task_idx], std::move(*result), previous_task);
 

diff --git a/src/Storages/MergeTree/ParallelReplicasReadingCoordinator.cpp b/src/Storages/MergeTree/ParallelReplicasReadingCoordinator.cpp
@@ -17,6 +17,7 @@
 #include <boost/algorithm/string/split.hpp>
 #include <fmt/core.h>
 #include <fmt/format.h>
+#include <Common/ThreadFuzzer.h>
 #include <Common/ElapsedTimeProfileEventIncrement.h>
 #include <Common/Exception.h>
 #include <Common/ProfileEvents.h>
@@ -428,6 +429,7 @@ void DefaultCoordinator::handleInitialAllRangesAnnouncement(InitialAllRangesAnno
 
     LOG_DEBUG(log, "Initial request from replica {}: {}", announcement.replica_num, announcement.describe());
 
+    ThreadFuzzer::maybeInjectSleep();
     initializeReadingState(std::move(announcement));
 
     if (replica_num >= stats.size())
@@ -740,6 +742,8 @@ ParallelReadResponse DefaultCoordinator::handleRequest(ParallelReadRequest reque
 {
     LOG_TRACE(log, "Handling request from replica {}, minimal marks size is {}", request.replica_num, request.min_number_of_marks);
 
+    ThreadFuzzer::maybeInjectSleep();
+
     ParallelReadResponse response;
 
     size_t current_mark_size = 0;