[Enhancement] show persistent index disk cost in be_tablets (backport #…

…35615) (#36280) Signed-off-by: luohaha <[email protected]>
StarRocks · Dec 2, 2023 · f13de63 · f13de63
1 parent 952da2d
commit f13de63
Show file tree

Hide file tree

Showing 5 changed files with 69 additions and 3 deletions.
diff --git a/be/src/exec/schema_scanner/schema_be_tablets_scanner.cpp b/be/src/exec/schema_scanner/schema_be_tablets_scanner.cpp
@@ -34,7 +34,7 @@ SchemaScanner::ColumnDesc SchemaBeTabletsScanner::_s_columns[] = {
         {"INDEX_MEM", TYPE_BIGINT, sizeof(int64_t), false},     {"CREATE_TIME", TYPE_BIGINT, sizeof(int64_t), false},
         {"STATE", TYPE_VARCHAR, sizeof(StringValue), false},    {"TYPE", TYPE_VARCHAR, sizeof(StringValue), false},
         {"DATA_DIR", TYPE_VARCHAR, sizeof(StringValue), false}, {"SHARD_ID", TYPE_BIGINT, sizeof(int64_t), false},
-        {"SCHEMA_HASH", TYPE_BIGINT, sizeof(int64_t), false},
+        {"SCHEMA_HASH", TYPE_BIGINT, sizeof(int64_t), false},   {"INDEX_DISK", TYPE_BIGINT, sizeof(int64_t), false},
 };
 
 SchemaBeTabletsScanner::SchemaBeTabletsScanner()
@@ -92,7 +92,7 @@ Status SchemaBeTabletsScanner::fill_chunk(ChunkPtr* chunk) {
     for (; _cur_idx < end; _cur_idx++) {
         auto& info = _infos[_cur_idx];
         for (const auto& [slot_id, index] : slot_id_to_index_map) {
-            if (slot_id < 1 || slot_id > 17) {
+            if (slot_id < 1 || slot_id > 18) {
                 return Status::InternalError(strings::Substitute("invalid slot id:$0", slot_id));
             }
             ColumnPtr column = (*chunk)->get_column_by_slot_id(slot_id);
@@ -170,18 +170,26 @@ Status SchemaBeTabletsScanner::fill_chunk(ChunkPtr* chunk) {
                 break;
             }
             case 15: {
+                // DATA_DIR
                 Slice type = Slice(info.data_dir);
                 fill_column_with_slot<TYPE_VARCHAR>(column.get(), (void*)&type);
                 break;
             }
             case 16: {
+                // SHARD_ID
                 fill_column_with_slot<TYPE_BIGINT>(column.get(), (void*)&info.shard_id);
                 break;
             }
             case 17: {
+                // SCHEMA_HASH
                 fill_column_with_slot<TYPE_BIGINT>(column.get(), (void*)&info.schema_hash);
                 break;
             }
+            case 18: {
+                // INDEX_DISK
+                fill_column_with_slot<TYPE_BIGINT>(column.get(), (void*)&info.index_disk_usage);
+                break;
+            }
             default:
                 break;
             }

diff --git a/be/src/exec/schema_scanner/schema_be_tablets_scanner.h b/be/src/exec/schema_scanner/schema_be_tablets_scanner.h
@@ -38,6 +38,7 @@ struct TabletBasicInfo {
     std::string data_dir;
     int64_t shard_id{0};
     int64_t schema_hash{0};
+    int64_t index_disk_usage{0};
 };
 
 class SchemaBeTabletsScanner : public SchemaScanner {

diff --git a/be/src/storage/tablet_updates.cpp b/be/src/storage/tablet_updates.cpp
@@ -14,8 +14,11 @@
 
 #include "storage/tablet_updates.h"
 
+#include <gutil/strings/util.h>
+
 #include <cmath>
 #include <ctime>
+#include <filesystem>
 #include <memory>
 
 #include "common/status.h"
@@ -2437,6 +2440,38 @@ size_t TabletUpdates::_get_rowset_num_deletes(const Rowset& rowset) {
     return num_dels;
 }
 
+Status TabletUpdates::_get_extra_file_size(int64_t* pindex_size, int64_t* col_size) {
+    const std::string tablet_path = _tablet.schema_hash_path();
+    try {
+        for (const auto& entry : std::filesystem::directory_iterator(tablet_path)) {
+            if (entry.is_regular_file()) {
+                std::string filename = entry.path().filename().string();
+
+                if (HasPrefixString(filename, "index.l")) {
+                    if (pindex_size != nullptr) {
+                        *pindex_size += std::filesystem::file_size(entry);
+                    }
+                } else if (HasSuffixString(filename, ".cols")) {
+                    // TODO skip the expired cols file
+                    if (col_size != nullptr) {
+                        *col_size += std::filesystem::file_size(entry);
+                    }
+                }
+            }
+        }
+    } catch (const std::filesystem::filesystem_error& ex) {
+        std::string err_msg = "Iterate dir " + tablet_path + " Filesystem error: " + ex.what();
+        return Status::InternalError(err_msg);
+    } catch (const std::exception& ex) {
+        std::string err_msg = "Iterate dir " + tablet_path + " Standard error: " + ex.what();
+        return Status::InternalError(err_msg);
+    } catch (...) {
+        std::string err_msg = "Iterate dir " + tablet_path + " Unknown exception occurred.";
+        return Status::InternalError(err_msg);
+    }
+    return Status::OK();
+}
+
 void TabletUpdates::get_tablet_info_extra(TTabletInfo* info) {
     int64_t min_readable_version = 0;
     int64_t max_readable_version = 0;
@@ -2480,13 +2515,22 @@ void TabletUpdates::get_tablet_info_extra(TTabletInfo* info) {
         LOG_EVERY_N(WARNING, 10) << "get_tablet_info_extra() some rowset stats not found tablet=" << _tablet.tablet_id()
                                  << " rowset=" << err_rowsets;
     }
+    int64_t pindex_size = 0;
+    int64_t col_size = 0;
+    Status st = _get_extra_file_size(&pindex_size, &col_size);
+    if (!st.ok()) {
+        // Ignore error status here, because we don't to break up tablet report because of get extra file size failure.
+        // So just print error log and keep going.
+        LOG(ERROR) << "get extra file size in primary table fail, tablet_id: " << _tablet.tablet_id()
+                   << " status: " << st;
+    }
     info->__set_version(version);
     info->__set_min_readable_version(min_readable_version);
     info->__set_max_readable_version(max_readable_version);
     info->__set_version_miss(has_pending);
     info->__set_version_count(version_count);
     info->__set_row_count(total_row);
-    info->__set_data_size(total_size);
+    info->__set_data_size(total_size + pindex_size + col_size);
     info->__set_is_error_state(_error);
 }
 
@@ -3469,6 +3513,16 @@ void TabletUpdates::get_basic_info_extra(TabletBasicInfo& info) {
         info.index_mem = index_entry->size();
         index_cache.release(index_entry);
     }
+    int64_t pindex_size = 0;
+    auto st = _get_extra_file_size(&pindex_size, nullptr);
+    if (!st.ok()) {
+        // Ignore error status here, because we don't to break up get basic info because of get pk index disk usage failure.
+        // So just print error log and keep going.
+        LOG(ERROR) << "get persistent index disk usage fail, tablet_id: " << _tablet.tablet_id()
+                   << ", error: " << st.get_error_msg();
+    } else {
+        info.index_disk_usage = pindex_size;
+    }
 }
 
 static double get_pk_index_write_amp_score_from_meta(Tablet* tablet) {

diff --git a/be/src/storage/tablet_updates.h b/be/src/storage/tablet_updates.h
@@ -411,6 +411,8 @@ class TabletUpdates {
 
     std::timed_mutex* get_index_lock() { return &_index_lock; }
 
+    Status _get_extra_file_size(int64_t* pindex_size, int64_t* col_size);
+
 private:
     Tablet& _tablet;
 

diff --git a/fe/fe-core/src/main/java/com/starrocks/catalog/system/information/BeTabletsSystemTable.java b/fe/fe-core/src/main/java/com/starrocks/catalog/system/information/BeTabletsSystemTable.java
@@ -46,6 +46,7 @@ public static SystemTable create() {
                         .column("DATA_DIR", ScalarType.createVarchar(NAME_CHAR_LEN))
                         .column("SHARD_ID", ScalarType.createType(PrimitiveType.BIGINT))
                         .column("SCHEMA_HASH", ScalarType.createType(PrimitiveType.BIGINT))
+                        .column("INDEX_DISK", ScalarType.createType(PrimitiveType.BIGINT))
                         .build(), TSchemaTableType.SCH_BE_TABLETS);
     }
 }