From c3e9bcd82188eaf5aa69adcd573197631353f0f3 Mon Sep 17 00:00:00 2001 From: Xinyi Zou Date: Wed, 29 Mar 2023 17:52:30 +0800 Subject: [PATCH 1/3] 1 --- be/src/common/config.h | 4 ---- be/src/common/daemon.cpp | 11 ----------- be/src/runtime/memory/mem_tracker_limiter.cpp | 5 +---- docs/en/community/developer-guide/be-vscode-dev.md | 2 ++ docs/zh-CN/community/developer-guide/be-vscode-dev.md | 2 ++ 5 files changed, 5 insertions(+), 19 deletions(-) diff --git a/be/src/common/config.h b/be/src/common/config.h index 7e3a5d9431c4ff..42baf9592c4449 100644 --- a/be/src/common/config.h +++ b/be/src/common/config.h @@ -77,10 +77,6 @@ CONF_Int64(max_sys_mem_available_low_water_mark_bytes, "1717986918"); // The size of the memory that gc wants to release each time, as a percentage of the mem limit. CONF_mString(process_minor_gc_size, "10%"); CONF_mString(process_full_gc_size, "20%"); -// Some caches have their own gc threads, such as segment cache. -// For caches that do not have a separate gc thread, perform regular gc in the memory maintenance thread. -// Currently only storage page cache, chunk allocator, more in the future. -CONF_mInt32(cache_gc_interval_s, "60"); // If true, when the process does not exceed the soft mem limit, the query memory will not be limited; // when the process memory exceeds the soft mem limit, the query with the largest ratio between the currently diff --git a/be/src/common/daemon.cpp b/be/src/common/daemon.cpp index c388fac9aafa05..176456f6369015 100644 --- a/be/src/common/daemon.cpp +++ b/be/src/common/daemon.cpp @@ -205,7 +205,6 @@ void Daemon::memory_maintenance_thread() { void Daemon::memory_gc_thread() { int32_t interval_milliseconds = config::memory_maintenance_sleep_time_ms; - int32_t cache_gc_interval_ms = config::cache_gc_interval_s * 1000; int32_t memory_minor_gc_sleep_time_ms = 0; int32_t memory_full_gc_sleep_time_ms = 0; int64_t cache_gc_freed_mem = 0; @@ -221,7 +220,6 @@ void Daemon::memory_gc_thread() { // No longer full gc and minor gc during sleep. memory_full_gc_sleep_time_ms = config::memory_gc_sleep_time_s * 1000; memory_minor_gc_sleep_time_ms = config::memory_gc_sleep_time_s * 1000; - cache_gc_interval_ms = config::cache_gc_interval_s * 1000; doris::MemTrackerLimiter::print_log_process_usage("process full gc", false); if (doris::MemInfo::process_full_gc()) { // If there is not enough memory to be gc, the process memory usage will not be printed in the next continuous gc. @@ -234,7 +232,6 @@ void Daemon::memory_gc_thread() { doris::MemInfo::soft_mem_limit())) { // No minor gc during sleep, but full gc is possible. memory_minor_gc_sleep_time_ms = config::memory_gc_sleep_time_s * 1000; - cache_gc_interval_ms = config::cache_gc_interval_s * 1000; doris::MemTrackerLimiter::print_log_process_usage("process minor gc", false); if (doris::MemInfo::process_minor_gc()) { doris::MemTrackerLimiter::enable_print_log_process_usage(); @@ -246,14 +243,6 @@ void Daemon::memory_gc_thread() { if (memory_minor_gc_sleep_time_ms > 0) { memory_minor_gc_sleep_time_ms -= interval_milliseconds; } - cache_gc_interval_ms -= interval_milliseconds; - if (cache_gc_interval_ms < 0) { - cache_gc_freed_mem = 0; - doris::MemInfo::process_cache_gc(cache_gc_freed_mem); - LOG(INFO) << fmt::format("Process regular GC Cache, Free Memory {} Bytes", - cache_gc_freed_mem); - cache_gc_interval_ms = config::cache_gc_interval_s * 1000; - } } } } diff --git a/be/src/runtime/memory/mem_tracker_limiter.cpp b/be/src/runtime/memory/mem_tracker_limiter.cpp index 9015485b8e2a81..f817da9d782df7 100644 --- a/be/src/runtime/memory/mem_tracker_limiter.cpp +++ b/be/src/runtime/memory/mem_tracker_limiter.cpp @@ -305,9 +305,6 @@ int64_t MemTrackerLimiter::free_top_memory_query(int64_t min_free_mem, std::lock_guard l(mem_tracker_limiter_pool[i].group_lock); for (auto tracker : mem_tracker_limiter_pool[i].trackers) { if (tracker->type() == type) { - if (tracker->consumption() <= 104857600) { // 100M small query does not cancel - continue; - } if (ExecEnv::GetInstance()->fragment_mgr()->query_is_canceled( label_to_queryid(tracker->label()))) { continue; @@ -351,7 +348,7 @@ int64_t MemTrackerLimiter::free_top_overcommit_query(int64_t min_free_mem, std::lock_guard l(mem_tracker_limiter_pool[i].group_lock); for (auto tracker : mem_tracker_limiter_pool[i].trackers) { if (tracker->type() == type) { - if (tracker->consumption() <= 104857600) { // 100M small query does not cancel + if (tracker->consumption() <= 33554432) { // 32M small query does not cancel continue; } if (ExecEnv::GetInstance()->fragment_mgr()->query_is_canceled( diff --git a/docs/en/community/developer-guide/be-vscode-dev.md b/docs/en/community/developer-guide/be-vscode-dev.md index 6552d5c8d78800..db02f6df3e2661 100644 --- a/docs/en/community/developer-guide/be-vscode-dev.md +++ b/docs/en/community/developer-guide/be-vscode-dev.md @@ -207,6 +207,8 @@ In the configuration **"request": "attach", "processId": PID**, these two config ps -ef | grep palo* ``` +Or write **"processId": "${command:pickProcess}"** to specify the pid when starting attach. + As shown in the figure: ![](/images/image-20210618095240216.png) diff --git a/docs/zh-CN/community/developer-guide/be-vscode-dev.md b/docs/zh-CN/community/developer-guide/be-vscode-dev.md index 4bcaf5d40730c9..338413dc388e5c 100644 --- a/docs/zh-CN/community/developer-guide/be-vscode-dev.md +++ b/docs/zh-CN/community/developer-guide/be-vscode-dev.md @@ -206,6 +206,8 @@ mkdir -p /soft/be/storage ps -ef | grep palo* ``` +或者写作 **"processId": "${command:pickProcess}"**，可在启动attach时指定pid. + 如图： ![](/images/image-20210618095240216.png) From 014c695fa0e0f6941dde209710749e3381eef9d4 Mon Sep 17 00:00:00 2001 From: Xinyi Zou Date: Wed, 29 Mar 2023 18:21:14 +0800 Subject: [PATCH 2/3] 2 --- bin/start_be.sh | 2 +- .../community/developer-guide/debug-tool.md | 31 ++++++++++--------- .../community/developer-guide/debug-tool.md | 11 ++++--- 3 files changed, 24 insertions(+), 20 deletions(-) diff --git a/bin/start_be.sh b/bin/start_be.sh index 199990a4b2a585..7204d6511441e4 100755 --- a/bin/start_be.sh +++ b/bin/start_be.sh @@ -236,7 +236,7 @@ if [[ -f "${DORIS_HOME}/conf/hdfs-site.xml" ]]; then fi # see https://github.com/apache/doris/blob/master/docs/zh-CN/community/developer-guide/debug-tool.md#jemalloc-heap-profile -export JEMALLOC_CONF="percpu_arena:percpu,background_thread:true,metadata_thp:auto,muzzy_decay_ms:30000,dirty_decay_ms:30000,oversize_threshold:0,lg_tcache_max:16,prof:true,prof_prefix:jeprof.out" +export JEMALLOC_CONF="percpu_arena:percpu,background_thread:true,metadata_thp:auto,muzzy_decay_ms:30000,dirty_decay_ms:30000,oversize_threshold:0,lg_tcache_max:16,prof_prefix:jeprof.out" if [[ "${RUN_DAEMON}" -eq 1 ]]; then nohup ${LIMIT:+${LIMIT}} "${DORIS_HOME}/lib/doris_be" "$@" >>"${LOG_DIR}/be.out" 2>&1 result.pdf ``` -默认heap profile采样间隔512K(2^19 B)，这会导致heap profile记录的内存通常只有10%，通过在`start_be.sh`中`JEMALLOC_CONF`增加`,lg_prof_sample:10`然后重启BE，可以减少采样间隔到1K (2^10 B), 更频繁的采样会使heap profile接近真实内存，但这会带来更大的性能损耗。详细参考 https://linux.die.net/man/3/jemalloc。 - -如果在做性能测试，尝试删掉`JEMALLOC_CONF`中的`,prof:true`，避免heap profile采样的性能损耗。 - ##### 3. heap dump by JEMALLOC_CONF 通过更改`start_be.sh` 中`JEMALLOC_CONF` 变量后重新启动BE 来进行heap dump From 63d8f1d75fc4517df45c82b89da2d43f8387d756 Mon Sep 17 00:00:00 2001 From: Xinyi Zou Date: Wed, 29 Mar 2023 18:28:08 +0800 Subject: [PATCH 3/3] 3 --- be/src/common/daemon.cpp | 1 - 1 file changed, 1 deletion(-) diff --git a/be/src/common/daemon.cpp b/be/src/common/daemon.cpp index 176456f6369015..9b7ee5dddc5a18 100644 --- a/be/src/common/daemon.cpp +++ b/be/src/common/daemon.cpp @@ -207,7 +207,6 @@ void Daemon::memory_gc_thread() { int32_t interval_milliseconds = config::memory_maintenance_sleep_time_ms; int32_t memory_minor_gc_sleep_time_ms = 0; int32_t memory_full_gc_sleep_time_ms = 0; - int64_t cache_gc_freed_mem = 0; while (!_stop_background_threads_latch.wait_for( std::chrono::milliseconds(interval_milliseconds))) { if (!MemInfo::initialized() || !ExecEnv::GetInstance()->initialized()) {