mito2/
metrics.rs

1// Copyright 2023 Greptime Team
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7//     http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15use std::time::Duration;
16
17use lazy_static::lazy_static;
18use prometheus::*;
19use puffin::puffin_manager::stager::StagerNotifier;
20
21/// Stage label.
22pub const STAGE_LABEL: &str = "stage";
23/// Type label.
24pub const TYPE_LABEL: &str = "type";
25const CACHE_EVICTION_CAUSE: &str = "cause";
26/// Reason to flush.
27pub const FLUSH_REASON: &str = "reason";
28/// File type label.
29pub const FILE_TYPE_LABEL: &str = "file_type";
30/// Region worker id label.
31pub const WORKER_LABEL: &str = "worker";
32/// Partition label.
33pub const PARTITION_LABEL: &str = "partition";
34/// Staging dir type label.
35pub const STAGING_TYPE: &str = "index_staging";
36/// Recycle bin type label.
37pub const RECYCLE_TYPE: &str = "recycle_bin";
38
39// Write metrics.
40lazy_static! {
41    /// Global write buffer size in bytes.
42    pub static ref WRITE_BUFFER_BYTES: IntGauge =
43        register_int_gauge!("greptime_mito_write_buffer_bytes", "mito write buffer bytes").unwrap();
44    /// Global memtable dictionary size in bytes.
45    pub static ref MEMTABLE_DICT_BYTES: IntGauge =
46        register_int_gauge!("greptime_mito_memtable_dict_bytes", "mito memtable dictionary size in bytes").unwrap();
47    /// Gauge for open regions in each worker.
48    pub static ref REGION_COUNT: IntGaugeVec =
49        register_int_gauge_vec!(
50            "greptime_mito_region_count",
51            "mito region count in each worker",
52            &[WORKER_LABEL],
53        ).unwrap();
54    /// Elapsed time to handle requests.
55    pub static ref HANDLE_REQUEST_ELAPSED: HistogramVec = register_histogram_vec!(
56            "greptime_mito_handle_request_elapsed",
57            "mito handle request elapsed",
58            &[TYPE_LABEL],
59            // 0.01 ~ 10000
60            exponential_buckets(0.01, 10.0, 7).unwrap(),
61        )
62        .unwrap();
63
64    // ------ Flush related metrics
65    /// Counter of scheduled flush requests.
66    /// Note that the flush scheduler may merge some flush requests.
67    pub static ref FLUSH_REQUESTS_TOTAL: IntCounterVec = register_int_counter_vec!(
68            "greptime_mito_flush_requests_total",
69            "mito flush requests total",
70            &[FLUSH_REASON]
71        )
72        .unwrap();
73    /// Counter of scheduled failed flush jobs.
74    pub static ref FLUSH_FAILURE_TOTAL: IntCounter =
75        register_int_counter!("greptime_mito_flush_failure_total", "mito flush failure total").unwrap();
76    /// Elapsed time of a flush job.
77    pub static ref FLUSH_ELAPSED: HistogramVec = register_histogram_vec!(
78            "greptime_mito_flush_elapsed",
79            "mito flush elapsed",
80            &[TYPE_LABEL],
81            // 1 ~ 625
82            exponential_buckets(1.0, 5.0, 6).unwrap(),
83        )
84        .unwrap();
85    /// Histogram of flushed bytes.
86    pub static ref FLUSH_BYTES_TOTAL: IntCounter =
87        register_int_counter!("greptime_mito_flush_bytes_total", "mito flush bytes total").unwrap();
88    /// Gauge for inflight flush tasks.
89    pub static ref INFLIGHT_FLUSH_COUNT: IntGauge =
90        register_int_gauge!(
91            "greptime_mito_inflight_flush_count",
92            "inflight flush count",
93        ).unwrap();
94    // ------ End of flush related metrics
95
96
97    // ------ Write related metrics
98    //
99    /// Counter of rejected write requests.
100    pub static ref WRITE_REJECT_TOTAL: IntCounter =
101        register_int_counter!("greptime_mito_write_reject_total", "mito write reject total").unwrap();
102    /// Elapsed time of each write stage.
103    pub static ref WRITE_STAGE_ELAPSED: HistogramVec = register_histogram_vec!(
104            "greptime_mito_write_stage_elapsed",
105            "mito write stage elapsed",
106            &[STAGE_LABEL],
107            // 0.01 ~ 1000
108            exponential_buckets(0.01, 10.0, 6).unwrap(),
109        )
110        .unwrap();
111    /// Counter of rows to write.
112    pub static ref WRITE_ROWS_TOTAL: IntCounterVec = register_int_counter_vec!(
113        "greptime_mito_write_rows_total",
114        "mito write rows total",
115        &[TYPE_LABEL]
116    )
117    .unwrap();
118}
119
120// Compaction metrics.
121lazy_static! {
122    /// Timer of different stages in compaction.
123    /// - pick
124    /// - merge (in parallel)
125    ///   - iter_source
126    ///   - write_batch
127    ///   - update_index
128    ///   - upload_parquet
129    ///   - upload puffin
130    /// - write_manifest
131    pub static ref COMPACTION_STAGE_ELAPSED: HistogramVec = register_histogram_vec!(
132        "greptime_mito_compaction_stage_elapsed",
133        "mito compaction stage elapsed",
134        &[STAGE_LABEL],
135        // 1 ~ 100000
136        exponential_buckets(1.0, 10.0, 6).unwrap(),
137    )
138    .unwrap();
139    /// Timer of whole compaction task.
140    pub static ref COMPACTION_ELAPSED_TOTAL: Histogram =
141        register_histogram!(
142        "greptime_mito_compaction_total_elapsed",
143        "mito compaction total elapsed",
144        // 1 ~ 100000
145        exponential_buckets(1.0, 10.0, 6).unwrap(),
146    ).unwrap();
147    /// Counter of all requested compaction task.
148    pub static ref COMPACTION_REQUEST_COUNT: IntCounter =
149        register_int_counter!("greptime_mito_compaction_requests_total", "mito compaction requests total").unwrap();
150    /// Counter of failed compaction task.
151    pub static ref COMPACTION_FAILURE_COUNT: IntCounter =
152        register_int_counter!("greptime_mito_compaction_failure_total", "mito compaction failure total").unwrap();
153
154    /// Gauge for inflight compaction tasks.
155    pub static ref INFLIGHT_COMPACTION_COUNT: IntGauge =
156        register_int_gauge!(
157            "greptime_mito_inflight_compaction_count",
158            "inflight compaction count",
159        ).unwrap();
160
161    /// Bytes reserved by compaction memory manager.
162    pub static ref COMPACTION_MEMORY_IN_USE: IntGauge =
163        register_int_gauge!(
164            "greptime_mito_compaction_memory_in_use_bytes",
165            "bytes currently reserved for compaction tasks",
166        )
167        .unwrap();
168    /// Configured compaction memory limit.
169    pub static ref COMPACTION_MEMORY_LIMIT: IntGauge =
170        register_int_gauge!(
171            "greptime_mito_compaction_memory_limit_bytes",
172            "maximum bytes allowed for compaction tasks",
173        )
174        .unwrap();
175    /// Wait time to obtain compaction memory.
176    pub static ref COMPACTION_MEMORY_WAIT: Histogram = register_histogram!(
177        "greptime_mito_compaction_memory_wait_seconds",
178        "time waiting for compaction memory",
179        // 0.01s ~ ~10s
180        exponential_buckets(0.01, 2.0, 10).unwrap(),
181    ).unwrap();
182    /// Counter of rejected compaction memory allocations.
183    pub static ref COMPACTION_MEMORY_REJECTED: IntCounterVec =
184        register_int_counter_vec!(
185            "greptime_mito_compaction_memory_rejected_total",
186            "number of compaction tasks rejected due to memory limit",
187            &[TYPE_LABEL]
188        ).unwrap();
189}
190
191// Query metrics.
192lazy_static! {
193    /// Timer of different stages in query.
194    pub static ref READ_STAGE_ELAPSED: HistogramVec = register_histogram_vec!(
195        "greptime_mito_read_stage_elapsed",
196        "mito read stage elapsed",
197        &[STAGE_LABEL],
198        // 0.01 ~ 10000
199        exponential_buckets(0.01, 10.0, 7).unwrap(),
200    )
201    .unwrap();
202    pub static ref READ_STAGE_FETCH_PAGES: Histogram = READ_STAGE_ELAPSED.with_label_values(&["fetch_pages"]);
203    /// Number of in-progress scan per partition.
204    pub static ref IN_PROGRESS_SCAN: IntGaugeVec = register_int_gauge_vec!(
205        "greptime_mito_in_progress_scan",
206        "mito in progress scan per partition",
207        &[TYPE_LABEL, PARTITION_LABEL]
208    )
209    .unwrap();
210    /// Counter of rows read from different source.
211    pub static ref READ_ROWS_TOTAL: IntCounterVec =
212        register_int_counter_vec!("greptime_mito_read_rows_total", "mito read rows total", &[TYPE_LABEL]).unwrap();
213    /// Counter of filtered rows during merge.
214    pub static ref MERGE_FILTER_ROWS_TOTAL: IntCounterVec =
215        register_int_counter_vec!("greptime_mito_merge_filter_rows_total", "mito merge filter rows total", &[TYPE_LABEL]).unwrap();
216    /// Counter of row groups read.
217    pub static ref READ_ROW_GROUPS_TOTAL: IntCounterVec =
218        register_int_counter_vec!("greptime_mito_read_row_groups_total", "mito read row groups total", &[TYPE_LABEL]).unwrap();
219    /// Counter of filtered rows by precise filter.
220    pub static ref PRECISE_FILTER_ROWS_TOTAL: IntCounterVec =
221        register_int_counter_vec!("greptime_mito_precise_filter_rows_total", "mito precise filter rows total", &[TYPE_LABEL]).unwrap();
222    pub static ref READ_ROWS_IN_ROW_GROUP_TOTAL: IntCounterVec =
223        register_int_counter_vec!("greptime_mito_read_rows_in_row_group_total", "mito read rows in row group total", &[TYPE_LABEL]).unwrap();
224    /// Histogram for the number of SSTs to scan per query.
225    pub static ref READ_SST_COUNT: Histogram = register_histogram!(
226        "greptime_mito_read_sst_count",
227        "Number of SSTs to scan in a scan task",
228        vec![1.0, 4.0, 8.0, 16.0, 32.0, 64.0, 256.0, 1024.0],
229    ).unwrap();
230    /// Histogram for the number of rows returned per query.
231    pub static ref READ_ROWS_RETURN: Histogram = register_histogram!(
232        "greptime_mito_read_rows_return",
233        "Number of rows returned in a scan task",
234        exponential_buckets(100.0, 10.0, 8).unwrap(),
235    ).unwrap();
236    /// Histogram for the number of batches returned per query.
237    pub static ref READ_BATCHES_RETURN: Histogram = register_histogram!(
238        "greptime_mito_read_batches_return",
239        "Number of rows returned in a scan task",
240        exponential_buckets(100.0, 10.0, 7).unwrap(),
241    ).unwrap();
242    /// Gauge for scan memory usage in bytes.
243    pub static ref SCAN_MEMORY_USAGE_BYTES: IntGauge = register_int_gauge!(
244        "greptime_mito_scan_memory_usage_bytes",
245        "current scan memory usage in bytes"
246    ).unwrap();
247    /// Counter of scan allocation attempts that could not acquire memory immediately.
248    pub static ref SCAN_MEMORY_EXHAUSTED_TOTAL: IntCounter = register_int_counter!(
249        "greptime_mito_scan_memory_exhausted_total",
250        "total number of times scan memory was unavailable for immediate acquisition"
251    ).unwrap();
252    /// Counter of scan requests that ultimately failed due to memory pressure.
253    pub static ref SCAN_REQUESTS_REJECTED_TOTAL: IntCounter = register_int_counter!(
254        "greptime_mito_scan_requests_rejected_total",
255        "total number of scan requests that ultimately failed due to memory limit"
256    ).unwrap();
257    /// Gauge for active file range builders in the pruner.
258    pub static ref PRUNER_ACTIVE_BUILDERS: IntGauge = register_int_gauge!(
259        "greptime_mito_pruner_active_builders",
260        "number of active file range builders in the pruner"
261    ).unwrap();
262}
263
264// Cache metrics.
265lazy_static! {
266    /// Cache hit counter.
267    pub static ref CACHE_HIT: IntCounterVec = register_int_counter_vec!(
268        "greptime_mito_cache_hit",
269        "mito cache hit",
270        &[TYPE_LABEL]
271    )
272    .unwrap();
273    /// Cache miss counter.
274    pub static ref CACHE_MISS: IntCounterVec = register_int_counter_vec!(
275        "greptime_mito_cache_miss",
276        "mito cache miss",
277        &[TYPE_LABEL]
278    )
279    .unwrap();
280    /// Cache size in bytes.
281    pub static ref CACHE_BYTES: IntGaugeVec = register_int_gauge_vec!(
282        "greptime_mito_cache_bytes",
283        "mito cache bytes",
284        &[TYPE_LABEL]
285    )
286    .unwrap();
287    /// Download bytes counter in the write cache.
288    pub static ref WRITE_CACHE_DOWNLOAD_BYTES_TOTAL: IntCounter = register_int_counter!(
289        "mito_write_cache_download_bytes_total",
290        "mito write cache download bytes total",
291    ).unwrap();
292    /// Timer of the downloading task in the write cache.
293    pub static ref WRITE_CACHE_DOWNLOAD_ELAPSED: HistogramVec = register_histogram_vec!(
294        "mito_write_cache_download_elapsed",
295        "mito write cache download elapsed",
296        &[TYPE_LABEL],
297        // 0.1 ~ 10000
298        exponential_buckets(0.1, 10.0, 6).unwrap(),
299    ).unwrap();
300    /// Number of inflight download tasks.
301    pub static ref WRITE_CACHE_INFLIGHT_DOWNLOAD: IntGauge = register_int_gauge!(
302        "mito_write_cache_inflight_download_count",
303        "mito write cache inflight download tasks",
304    ).unwrap();
305    /// Upload bytes counter.
306    pub static ref UPLOAD_BYTES_TOTAL: IntCounter = register_int_counter!(
307        "mito_upload_bytes_total",
308        "mito upload bytes total",
309    )
310    .unwrap();
311    /// Cache eviction counter, labeled with cache type and eviction reason.
312    pub static ref CACHE_EVICTION: IntCounterVec = register_int_counter_vec!(
313        "greptime_mito_cache_eviction",
314        "mito cache eviction",
315        &[TYPE_LABEL, CACHE_EVICTION_CAUSE]
316    ).unwrap();
317}
318
319// Index metrics.
320lazy_static! {
321    // Index metrics.
322    /// Timer of index application.
323    pub static ref INDEX_APPLY_ELAPSED: HistogramVec = register_histogram_vec!(
324        "greptime_index_apply_elapsed",
325        "index apply elapsed",
326        &[TYPE_LABEL],
327        // 0.01 ~ 1000
328        exponential_buckets(0.01, 10.0, 6).unwrap(),
329    )
330    .unwrap();
331    /// Gauge of index apply memory usage.
332    pub static ref INDEX_APPLY_MEMORY_USAGE: IntGauge = register_int_gauge!(
333        "greptime_index_apply_memory_usage",
334        "index apply memory usage",
335    )
336    .unwrap();
337    /// Timer of index creation.
338    pub static ref INDEX_CREATE_ELAPSED: HistogramVec = register_histogram_vec!(
339        "greptime_index_create_elapsed",
340        "index create elapsed",
341        &[STAGE_LABEL, TYPE_LABEL],
342        // 0.1 ~ 10000
343        exponential_buckets(0.1, 10.0, 6).unwrap(),
344    )
345    .unwrap();
346    /// Counter of rows indexed.
347    pub static ref INDEX_CREATE_ROWS_TOTAL: IntCounterVec = register_int_counter_vec!(
348        "greptime_index_create_rows_total",
349        "index create rows total",
350        &[TYPE_LABEL],
351    )
352    .unwrap();
353    /// Counter of created index bytes.
354    pub static ref INDEX_CREATE_BYTES_TOTAL: IntCounterVec = register_int_counter_vec!(
355        "greptime_index_create_bytes_total",
356        "index create bytes total",
357        &[TYPE_LABEL],
358    )
359    .unwrap();
360    /// Gauge of index create memory usage.
361    pub static ref INDEX_CREATE_MEMORY_USAGE: IntGaugeVec = register_int_gauge_vec!(
362        "greptime_index_create_memory_usage",
363        "index create memory usage",
364        &[TYPE_LABEL],
365    ).unwrap();
366    /// Counter of r/w bytes on index related IO operations.
367    pub static ref INDEX_IO_BYTES_TOTAL: IntCounterVec = register_int_counter_vec!(
368        "greptime_index_io_bytes_total",
369        "index io bytes total",
370        &[TYPE_LABEL, FILE_TYPE_LABEL]
371    )
372    .unwrap();
373    /// Counter of read bytes on puffin files.
374    pub static ref INDEX_PUFFIN_READ_BYTES_TOTAL: IntCounter = INDEX_IO_BYTES_TOTAL
375        .with_label_values(&["read", "puffin"]);
376    /// Counter of write bytes on puffin files.
377    pub static ref INDEX_PUFFIN_WRITE_BYTES_TOTAL: IntCounter = INDEX_IO_BYTES_TOTAL
378        .with_label_values(&["write", "puffin"]);
379    /// Counter of read bytes on intermediate files.
380    pub static ref INDEX_INTERMEDIATE_READ_BYTES_TOTAL: IntCounter = INDEX_IO_BYTES_TOTAL
381        .with_label_values(&["read", "intermediate"]);
382    /// Counter of write bytes on intermediate files.
383    pub static ref INDEX_INTERMEDIATE_WRITE_BYTES_TOTAL: IntCounter = INDEX_IO_BYTES_TOTAL
384        .with_label_values(&["write", "intermediate"]);
385
386    /// Counter of r/w operations on index related IO operations, e.g. read, write, seek and flush.
387    pub static ref INDEX_IO_OP_TOTAL: IntCounterVec = register_int_counter_vec!(
388        "greptime_index_io_op_total",
389        "index io op total",
390        &[TYPE_LABEL, FILE_TYPE_LABEL]
391    )
392    .unwrap();
393    /// Counter of read operations on puffin files.
394    pub static ref INDEX_PUFFIN_READ_OP_TOTAL: IntCounter = INDEX_IO_OP_TOTAL
395        .with_label_values(&["read", "puffin"]);
396    /// Counter of seek operations on puffin files.
397    pub static ref INDEX_PUFFIN_SEEK_OP_TOTAL: IntCounter = INDEX_IO_OP_TOTAL
398        .with_label_values(&["seek", "puffin"]);
399    /// Counter of write operations on puffin files.
400    pub static ref INDEX_PUFFIN_WRITE_OP_TOTAL: IntCounter = INDEX_IO_OP_TOTAL
401        .with_label_values(&["write", "puffin"]);
402    /// Counter of flush operations on puffin files.
403    pub static ref INDEX_PUFFIN_FLUSH_OP_TOTAL: IntCounter = INDEX_IO_OP_TOTAL
404        .with_label_values(&["flush", "puffin"]);
405    /// Counter of read operations on intermediate files.
406    pub static ref INDEX_INTERMEDIATE_READ_OP_TOTAL: IntCounter = INDEX_IO_OP_TOTAL
407        .with_label_values(&["read", "intermediate"]);
408    /// Counter of seek operations on intermediate files.
409    pub static ref INDEX_INTERMEDIATE_SEEK_OP_TOTAL: IntCounter = INDEX_IO_OP_TOTAL
410        .with_label_values(&["seek", "intermediate"]);
411    /// Counter of write operations on intermediate files.
412    pub static ref INDEX_INTERMEDIATE_WRITE_OP_TOTAL: IntCounter = INDEX_IO_OP_TOTAL
413        .with_label_values(&["write", "intermediate"]);
414    /// Counter of flush operations on intermediate files.
415    pub static ref INDEX_INTERMEDIATE_FLUSH_OP_TOTAL: IntCounter = INDEX_IO_OP_TOTAL
416        .with_label_values(&["flush", "intermediate"]);
417}
418
419lazy_static! {
420    /// Partition tree memtable data buffer freeze metrics
421    pub static ref PARTITION_TREE_DATA_BUFFER_FREEZE_STAGE_ELAPSED: HistogramVec = register_histogram_vec!(
422        "greptime_partition_tree_buffer_freeze_stage_elapsed",
423        "mito partition tree data buffer freeze stage elapsed",
424        &[STAGE_LABEL],
425        // 0.01 ~ 1000
426        exponential_buckets(0.01, 10.0, 6).unwrap(),
427    )
428    .unwrap();
429
430    /// Partition tree memtable read path metrics
431    pub static ref PARTITION_TREE_READ_STAGE_ELAPSED: HistogramVec = register_histogram_vec!(
432        "greptime_partition_tree_read_stage_elapsed",
433        "mito partition tree read stage elapsed",
434        &[STAGE_LABEL],
435        // 0.01 ~ 1000
436        exponential_buckets(0.01, 10.0, 6).unwrap(),
437    )
438    .unwrap();
439
440    // ------- End of partition tree memtable metrics.
441
442
443    // Manifest related metrics:
444
445    /// Elapsed time of manifest operation. Labeled with "op".
446    pub static ref MANIFEST_OP_ELAPSED: HistogramVec = register_histogram_vec!(
447        "greptime_manifest_op_elapsed",
448        "mito manifest operation elapsed",
449        &["op"],
450        // 0.01 ~ 1000
451        exponential_buckets(0.01, 10.0, 6).unwrap(),
452    ).unwrap();
453
454
455    pub static ref REGION_WORKER_HANDLE_WRITE_ELAPSED: HistogramVec = register_histogram_vec!(
456        "greptime_region_worker_handle_write",
457        "elapsed time for handling writes in region worker loop",
458        &["stage"],
459        exponential_buckets(0.001, 10.0, 5).unwrap()
460    ).unwrap();
461
462}
463
464lazy_static! {
465    /// Counter for compaction input file size.
466    pub static ref COMPACTION_INPUT_BYTES: Counter = register_counter!(
467        "greptime_mito_compaction_input_bytes",
468        "mito compaction input file size",
469        ).unwrap();
470
471    /// Counter for compaction output file size.
472    pub static ref COMPACTION_OUTPUT_BYTES: Counter = register_counter!(
473        "greptime_mito_compaction_output_bytes",
474        "mito compaction output file size",
475        ).unwrap();
476
477    /// Active series count in TimeSeriesMemtable
478    pub static ref MEMTABLE_ACTIVE_SERIES_COUNT: IntGauge = register_int_gauge!(
479        "greptime_mito_memtable_active_series_count",
480        "active time series count in TimeSeriesMemtable",
481        ).unwrap();
482
483    /// Active field builder count in TimeSeriesMemtable
484    pub static ref MEMTABLE_ACTIVE_FIELD_BUILDER_COUNT: IntGauge = register_int_gauge!(
485        "greptime_mito_memtable_field_builder_count",
486        "active field builder count in TimeSeriesMemtable",
487        ).unwrap();
488
489    /// Number of stalling write requests in each worker.
490    pub static ref WRITE_STALLING: IntGaugeVec = register_int_gauge_vec!(
491            "greptime_mito_write_stalling_count",
492            "mito stalled write request in each worker",
493            &[WORKER_LABEL]
494        ).unwrap();
495    /// Number of ref files
496    pub static ref GC_REF_FILE_CNT: IntGauge = register_int_gauge!(
497            "greptime_gc_ref_file_count",
498            "gc ref file count",
499        ).unwrap();
500    /// Total number of stalled write requests.
501    pub static ref WRITE_STALL_TOTAL: IntCounter = register_int_counter!(
502        "greptime_mito_write_stall_total",
503        "Total number of stalled write requests"
504    ).unwrap();
505    /// Time waiting for requests to be handled by the region worker.
506    pub static ref REQUEST_WAIT_TIME: HistogramVec = register_histogram_vec!(
507            "greptime_mito_request_wait_time",
508            "mito request wait time before being handled by region worker",
509            &[WORKER_LABEL],
510            // 0.001 ~ 10000
511            exponential_buckets(0.001, 10.0, 8).unwrap(),
512        )
513        .unwrap();
514
515    /// Counter for the number of files deleted by the GC worker.
516    pub static ref GC_DELETE_FILE_CNT: IntCounter =
517        register_int_counter!(
518            "greptime_mito_gc_delete_file_count",
519            "mito gc deleted file count",
520        ).unwrap();
521
522    /// Counter for the number of unparsable files skipped by GC.
523    pub static ref GC_SKIPPED_UNPARSABLE_FILES: IntCounter =
524        register_int_counter!(
525            "greptime_mito_gc_skipped_unparsable_files",
526            "mito gc skipped unparsable files count",
527        ).unwrap();
528
529    /// Counter for the number of orphaned index files found by GC.
530    pub static ref GC_ORPHANED_INDEX_FILES: IntCounter =
531        register_int_counter!(
532            "greptime_mito_gc_orphaned_index_files",
533            "mito gc orphaned index files count",
534        ).unwrap();
535
536    /// Histogram for GC operation duration by stage.
537    pub static ref GC_DURATION_SECONDS: HistogramVec = register_histogram_vec!(
538        "greptime_mito_gc_duration_seconds",
539        "GC operation duration by stage",
540        &[STAGE_LABEL],
541        exponential_buckets(0.01, 10.0, 6).unwrap(),
542    ).unwrap();
543
544    /// Counter for GC runs by mode.
545    pub static ref GC_RUNS_TOTAL: IntCounterVec = register_int_counter_vec!(
546        "greptime_mito_gc_runs_total",
547        "Total GC runs by mode",
548        &["mode"],
549    ).unwrap();
550
551    /// Counter for GC errors by type.
552    pub static ref GC_ERRORS_TOTAL: IntCounterVec = register_int_counter_vec!(
553        "greptime_mito_gc_errors_total",
554        "Total GC errors by type",
555        &["error_type"],
556    ).unwrap();
557
558    /// Counter for total files deleted by GC, labeled by file type.
559    pub static ref GC_FILES_DELETED_TOTAL: IntCounterVec = register_int_counter_vec!(
560        "greptime_mito_gc_files_deleted_total",
561        "Total files deleted by GC",
562        &[FILE_TYPE_LABEL],
563    ).unwrap();
564
565    /// Total number of files downloaded during cache fill on region open.
566    pub static ref CACHE_FILL_DOWNLOADED_FILES: IntCounter = register_int_counter!(
567        "mito_cache_fill_downloaded_files",
568        "mito cache fill downloaded files count",
569    ).unwrap();
570
571    /// Number of files pending download during cache fill on region open.
572    pub static ref CACHE_FILL_PENDING_FILES: IntGauge = register_int_gauge!(
573        "mito_cache_fill_pending_files",
574        "mito cache fill pending files count",
575    ).unwrap();
576
577    /// Counter of flush files.
578    pub static ref FLUSH_FILE_TOTAL: IntCounter =
579        register_int_counter!("greptime_mito_flush_file_total", "mito flushed file count").unwrap();
580}
581
582/// Stager notifier to collect metrics.
583pub struct StagerMetrics {
584    cache_hit: IntCounter,
585    cache_miss: IntCounter,
586    staging_cache_bytes: IntGauge,
587    recycle_cache_bytes: IntGauge,
588    cache_eviction: IntCounter,
589    staging_miss_read: Histogram,
590}
591
592impl StagerMetrics {
593    /// Creates a new stager notifier.
594    pub fn new() -> Self {
595        Self {
596            cache_hit: CACHE_HIT.with_label_values(&[STAGING_TYPE]),
597            cache_miss: CACHE_MISS.with_label_values(&[STAGING_TYPE]),
598            staging_cache_bytes: CACHE_BYTES.with_label_values(&[STAGING_TYPE]),
599            recycle_cache_bytes: CACHE_BYTES.with_label_values(&[RECYCLE_TYPE]),
600            cache_eviction: CACHE_EVICTION.with_label_values(&[STAGING_TYPE, "size"]),
601            staging_miss_read: READ_STAGE_ELAPSED.with_label_values(&["staging_miss_read"]),
602        }
603    }
604}
605
606impl Default for StagerMetrics {
607    fn default() -> Self {
608        Self::new()
609    }
610}
611
612impl StagerNotifier for StagerMetrics {
613    fn on_cache_hit(&self, _size: u64) {
614        self.cache_hit.inc();
615    }
616
617    fn on_cache_miss(&self, _size: u64) {
618        self.cache_miss.inc();
619    }
620
621    fn on_cache_insert(&self, size: u64) {
622        self.staging_cache_bytes.add(size as i64);
623    }
624
625    fn on_load_dir(&self, duration: Duration) {
626        self.staging_miss_read.observe(duration.as_secs_f64());
627    }
628
629    fn on_load_blob(&self, duration: Duration) {
630        self.staging_miss_read.observe(duration.as_secs_f64());
631    }
632
633    fn on_cache_evict(&self, size: u64) {
634        self.cache_eviction.inc();
635        self.staging_cache_bytes.sub(size as i64);
636    }
637
638    fn on_recycle_insert(&self, size: u64) {
639        self.recycle_cache_bytes.add(size as i64);
640    }
641
642    fn on_recycle_clear(&self, size: u64) {
643        self.recycle_cache_bytes.sub(size as i64);
644    }
645}
mito2/metrics.rs

mito2/
metrics.rs
