risingwave_storage/
opts.rs

1// Copyright 2023 RisingWave Labs
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 risingwave_common::config::streaming::CacheRefillPolicy;
16use risingwave_common::config::{
17    EvictionConfig, ObjectStoreConfig, RwConfig, StorageMemoryConfig, extract_storage_memory_config,
18};
19use risingwave_common::system_param::reader::{SystemParamsRead, SystemParamsReader};
20use risingwave_common::system_param::system_params_for_test;
21
22#[derive(Clone, Debug)]
23pub struct StorageOpts {
24    /// The size of parallel task for one compact/flush job.
25    pub parallel_compact_size_mb: u32,
26    /// Target size of the Sstable.
27    pub sstable_size_mb: u32,
28    /// Minimal target size of the Sstable to store data of different state-table in independent files as soon as possible.
29    pub min_sstable_size_mb: u32,
30    /// Size of each block in bytes in SST.
31    pub block_size_kb: u32,
32    /// Deprecated and ignored by SST filter builders; kept for backward compatibility.
33    pub bloom_false_positive: f64,
34    /// parallelism while syncing share buffers into L0 SST. Should NOT be 0.
35    pub share_buffers_sync_parallelism: u32,
36    /// Worker threads number of dedicated tokio runtime for share buffer compaction. 0 means use
37    /// tokio's default value (number of CPU core).
38    pub share_buffer_compaction_worker_threads_number: u32,
39    /// Maximum shared buffer size, writes attempting to exceed the capacity will stall until there
40    /// is enough space.
41    pub shared_buffer_capacity_mb: usize,
42    /// The shared buffer will start flushing data to object when the ratio of memory usage to the
43    /// shared buffer capacity exceed such ratio.
44    pub shared_buffer_flush_ratio: f32,
45    /// The minimum total flush size of shared buffer spill. When a shared buffer spill is trigger,
46    /// the total flush size across multiple epochs should be at least higher than this size.
47    pub shared_buffer_min_batch_flush_size_mb: usize,
48    /// Remote directory for storing data and metadata objects.
49    pub data_directory: String,
50    /// Whether to enable write conflict detection
51    pub write_conflict_detection_enabled: bool,
52    /// Capacity of sstable block cache.
53    pub block_cache_capacity_mb: usize,
54    /// the number of block-cache shard. Less shard means that more concurrent-conflict.
55    pub block_cache_shard_num: usize,
56    /// Eviction config for block cache.
57    pub block_cache_eviction_config: EvictionConfig,
58    /// Capacity of sstable meta cache.
59    pub meta_cache_capacity_mb: usize,
60    /// the number of meta-cache shard. Less shard means that more concurrent-conflict.
61    pub meta_cache_shard_num: usize,
62    /// Eviction config for meta cache.
63    pub meta_cache_eviction_config: EvictionConfig,
64    /// max memory usage for large query.
65    pub prefetch_buffer_capacity_mb: usize,
66
67    pub max_cached_recent_versions_number: usize,
68
69    pub max_prefetch_block_number: usize,
70
71    pub disable_remote_compactor: bool,
72    /// Number of tasks shared buffer can upload in parallel.
73    pub share_buffer_upload_concurrency: usize,
74    /// Capacity of sstable meta cache.
75    pub compactor_memory_limit_mb: usize,
76    /// compactor streaming iterator recreate timeout.
77    /// deprecated
78    pub compact_iter_recreate_timeout_ms: u64,
79    /// Number of SST ids fetched from meta per RPC
80    pub sstable_id_remote_fetch_number: u32,
81    /// Whether to enable streaming upload for sstable.
82    pub min_sst_size_for_streaming_upload: u64,
83    pub max_concurrent_compaction_task_number: u64,
84    pub max_version_pinning_duration_sec: u64,
85    pub compactor_iter_max_io_retry_times: usize,
86
87    /// If set, block metadata keys will be shortened when their length exceeds this threshold.
88    pub shorten_block_meta_key_threshold: Option<usize>,
89
90    pub data_file_cache_dir: String,
91    pub data_file_cache_capacity_mb: usize,
92    pub data_file_cache_file_capacity_mb: usize,
93    pub data_file_cache_flushers: usize,
94    pub data_file_cache_reclaimers: usize,
95    pub data_file_cache_recover_mode: foyer::RecoverMode,
96    pub data_file_cache_recover_concurrency: usize,
97    pub data_file_cache_indexer_shards: usize,
98    pub data_file_cache_compression: foyer::Compression,
99    pub data_file_cache_flush_buffer_threshold_mb: usize,
100    pub data_file_cache_submit_queue_size_threshold_mb: usize,
101    pub data_file_cache_fifo_probation_ratio: f64,
102    pub data_file_cache_blob_index_size_kb: usize,
103    pub data_file_cache_runtime_config: foyer::RuntimeOptions,
104    pub data_file_cache_throttle: foyer::Throttle,
105
106    pub cache_refill_data_refill_levels: Vec<u32>,
107    pub cache_refill_timeout_ms: u64,
108    pub cache_refill_meta_refill_concurrency: usize,
109    pub cache_refill_concurrency: usize,
110    pub cache_refill_recent_filter_shards: usize,
111    pub cache_refill_recent_filter_layers: usize,
112    pub cache_refill_recent_filter_rotate_interval_ms: usize,
113    pub cache_refill_unit: usize,
114    pub cache_refill_threshold: f64,
115    pub cache_refill_skip_recent_filter: bool,
116    pub cache_refill_skip_inheritance_filter: bool,
117    pub cache_refill_table_cache_refill_default_policy: CacheRefillPolicy,
118
119    pub meta_file_cache_dir: String,
120    pub meta_file_cache_capacity_mb: usize,
121    pub meta_file_cache_file_capacity_mb: usize,
122    pub meta_file_cache_flushers: usize,
123    pub meta_file_cache_reclaimers: usize,
124    pub meta_file_cache_recover_mode: foyer::RecoverMode,
125    pub meta_file_cache_recover_concurrency: usize,
126    pub meta_file_cache_indexer_shards: usize,
127    pub meta_file_cache_compression: foyer::Compression,
128    pub meta_file_cache_flush_buffer_threshold_mb: usize,
129    pub meta_file_cache_submit_queue_size_threshold_mb: usize,
130    pub meta_file_cache_fifo_probation_ratio: f64,
131    pub meta_file_cache_blob_index_size_kb: usize,
132    pub meta_file_cache_runtime_config: foyer::RuntimeOptions,
133    pub meta_file_cache_throttle: foyer::Throttle,
134    pub sst_skip_bloom_filter_in_serde: bool,
135
136    pub vector_file_block_size_kb: usize,
137    pub vector_block_cache_capacity_mb: usize,
138    pub vector_block_cache_shard_num: usize,
139    pub vector_block_cache_eviction_config: EvictionConfig,
140    pub vector_meta_cache_capacity_mb: usize,
141    pub vector_meta_cache_shard_num: usize,
142    pub vector_meta_cache_eviction_config: EvictionConfig,
143
144    /// The storage url for storing backups.
145    pub backup_storage_url: String,
146    /// The storage directory for storing backups.
147    pub backup_storage_directory: String,
148    /// max time which wait for preload. 0 represent do not do any preload.
149    pub max_preload_wait_time_mill: u64,
150
151    pub compactor_max_sst_key_count: u64,
152    pub compactor_max_task_multiplier: f32,
153    pub compactor_max_sst_size: u64,
154    /// enable `FastCompactorRunner`.
155    pub enable_fast_compaction: bool,
156    pub check_compaction_result: bool,
157    pub max_preload_io_retry_times: usize,
158    pub compactor_fast_max_compact_delete_ratio: u32,
159    pub compactor_fast_max_compact_task_size: u64,
160
161    pub mem_table_spill_threshold: usize,
162
163    pub compactor_concurrent_uploading_sst_count: Option<usize>,
164
165    pub compactor_max_overlap_sst_count: usize,
166
167    /// The maximum number of meta files that can be preloaded.
168    pub compactor_max_preload_meta_file_count: usize,
169
170    pub object_store_config: ObjectStoreConfig,
171    pub time_travel_version_cache_capacity: u64,
172    pub table_change_log_cache_capacity: u64,
173
174    pub iceberg_compaction_enable_validate: bool,
175    pub iceberg_compaction_max_record_batch_rows: usize,
176    pub iceberg_compaction_write_parquet_max_row_group_rows: usize,
177    pub iceberg_compaction_min_size_per_partition_mb: u32,
178    pub iceberg_compaction_max_file_count_per_partition: u32,
179    pub iceberg_compaction_target_binpack_group_size_mb: Option<u64>,
180    pub iceberg_compaction_min_group_size_mb: Option<u64>,
181    pub iceberg_compaction_min_group_file_count: Option<usize>,
182
183    /// The ratio of iceberg compaction max parallelism to the number of CPU cores
184    pub iceberg_compaction_task_parallelism_ratio: f32,
185    /// Whether to enable heuristic output parallelism in iceberg compaction.
186    pub iceberg_compaction_enable_heuristic_output_parallelism: bool,
187    /// Maximum number of concurrent file close operations
188    pub iceberg_compaction_max_concurrent_closes: usize,
189    /// Whether to enable dynamic size estimation for iceberg compaction.
190    pub iceberg_compaction_enable_dynamic_size_estimation: bool,
191    /// The smoothing factor for size estimation in iceberg compaction.(default: 0.3)
192    pub iceberg_compaction_size_estimation_smoothing_factor: f64,
193    /// Multiplier for pending waiting parallelism budget for iceberg compaction task queue.
194    pub iceberg_compaction_pending_parallelism_budget_multiplier: f32,
195    /// Pull interval for iceberg compaction task requests in milliseconds.
196    pub iceberg_compaction_pull_interval_ms: u64,
197    /// Whether to enable prefetch for iceberg compaction.
198    pub iceberg_compaction_enable_prefetch: bool,
199}
200
201impl Default for StorageOpts {
202    fn default() -> Self {
203        let c = RwConfig::default();
204        let p = system_params_for_test();
205        let s = extract_storage_memory_config(&c);
206        Self::from((&c, &p.into(), &s))
207    }
208}
209
210impl From<(&RwConfig, &SystemParamsReader, &StorageMemoryConfig)> for StorageOpts {
211    fn from((c, p, s): (&RwConfig, &SystemParamsReader, &StorageMemoryConfig)) -> Self {
212        let mut data_file_cache_throttle = c.storage.data_file_cache.throttle.clone();
213        if data_file_cache_throttle.write_throughput.is_none() {
214            data_file_cache_throttle = data_file_cache_throttle.with_write_throughput(
215                c.storage.data_file_cache.insert_rate_limit_mb * 1024 * 1024,
216            );
217        }
218        let mut meta_file_cache_throttle = c.storage.meta_file_cache.throttle.clone();
219        if meta_file_cache_throttle.write_throughput.is_none() {
220            meta_file_cache_throttle = meta_file_cache_throttle.with_write_throughput(
221                c.storage.meta_file_cache.insert_rate_limit_mb * 1024 * 1024,
222            );
223        }
224
225        Self {
226            parallel_compact_size_mb: p.parallel_compact_size_mb(),
227            sstable_size_mb: p.sstable_size_mb(),
228            min_sstable_size_mb: c.storage.min_sstable_size_mb,
229            block_size_kb: p.block_size_kb(),
230            bloom_false_positive: p.bloom_false_positive(),
231            share_buffers_sync_parallelism: c.storage.share_buffers_sync_parallelism,
232            share_buffer_compaction_worker_threads_number: c
233                .storage
234                .share_buffer_compaction_worker_threads_number,
235            shared_buffer_capacity_mb: s.shared_buffer_capacity_mb,
236            shared_buffer_flush_ratio: c.storage.shared_buffer_flush_ratio,
237            shared_buffer_min_batch_flush_size_mb: c.storage.shared_buffer_min_batch_flush_size_mb,
238            data_directory: p.data_directory().to_owned(),
239            write_conflict_detection_enabled: c.storage.write_conflict_detection_enabled,
240            block_cache_capacity_mb: s.block_cache_capacity_mb,
241            block_cache_shard_num: s.block_cache_shard_num,
242            block_cache_eviction_config: s.block_cache_eviction_config.clone(),
243            meta_cache_capacity_mb: s.meta_cache_capacity_mb,
244            meta_cache_shard_num: s.meta_cache_shard_num,
245            meta_cache_eviction_config: s.meta_cache_eviction_config.clone(),
246            prefetch_buffer_capacity_mb: s.prefetch_buffer_capacity_mb,
247            max_cached_recent_versions_number: c.storage.max_cached_recent_versions_number,
248            max_prefetch_block_number: c.storage.max_prefetch_block_number,
249            disable_remote_compactor: c.storage.disable_remote_compactor,
250            share_buffer_upload_concurrency: c.storage.share_buffer_upload_concurrency,
251            compactor_memory_limit_mb: s.compactor_memory_limit_mb,
252            sstable_id_remote_fetch_number: c.storage.sstable_id_remote_fetch_number,
253            min_sst_size_for_streaming_upload: c.storage.min_sst_size_for_streaming_upload,
254            max_concurrent_compaction_task_number: c.storage.max_concurrent_compaction_task_number,
255            max_version_pinning_duration_sec: c.storage.max_version_pinning_duration_sec,
256            data_file_cache_dir: c.storage.data_file_cache.dir.clone(),
257            data_file_cache_capacity_mb: c.storage.data_file_cache.capacity_mb,
258            data_file_cache_file_capacity_mb: c.storage.data_file_cache.file_capacity_mb,
259            data_file_cache_flushers: c.storage.data_file_cache.flushers,
260            data_file_cache_reclaimers: c.storage.data_file_cache.reclaimers,
261            data_file_cache_recover_mode: c.storage.data_file_cache.recover_mode,
262            data_file_cache_recover_concurrency: c.storage.data_file_cache.recover_concurrency,
263            data_file_cache_indexer_shards: c.storage.data_file_cache.indexer_shards,
264            data_file_cache_compression: c.storage.data_file_cache.compression,
265            data_file_cache_flush_buffer_threshold_mb: s.block_file_cache_flush_buffer_threshold_mb,
266            data_file_cache_submit_queue_size_threshold_mb: c
267                .storage
268                .data_file_cache
269                .submit_queue_size_threshold_mb,
270            data_file_cache_fifo_probation_ratio: c.storage.data_file_cache.fifo_probation_ratio,
271            data_file_cache_blob_index_size_kb: c.storage.data_file_cache.blob_index_size_kb,
272            data_file_cache_runtime_config: c.storage.data_file_cache.runtime_config.clone(),
273            data_file_cache_throttle,
274            meta_file_cache_dir: c.storage.meta_file_cache.dir.clone(),
275            meta_file_cache_capacity_mb: c.storage.meta_file_cache.capacity_mb,
276            meta_file_cache_file_capacity_mb: c.storage.meta_file_cache.file_capacity_mb,
277            meta_file_cache_flushers: c.storage.meta_file_cache.flushers,
278            meta_file_cache_reclaimers: c.storage.meta_file_cache.reclaimers,
279            meta_file_cache_recover_mode: c.storage.meta_file_cache.recover_mode,
280            meta_file_cache_recover_concurrency: c.storage.meta_file_cache.recover_concurrency,
281            meta_file_cache_indexer_shards: c.storage.meta_file_cache.indexer_shards,
282            meta_file_cache_compression: c.storage.meta_file_cache.compression,
283            meta_file_cache_flush_buffer_threshold_mb: s.meta_file_cache_flush_buffer_threshold_mb,
284            meta_file_cache_submit_queue_size_threshold_mb: c
285                .storage
286                .meta_file_cache
287                .submit_queue_size_threshold_mb,
288            meta_file_cache_fifo_probation_ratio: c.storage.meta_file_cache.fifo_probation_ratio,
289            meta_file_cache_blob_index_size_kb: c.storage.meta_file_cache.blob_index_size_kb,
290            meta_file_cache_runtime_config: c.storage.meta_file_cache.runtime_config.clone(),
291            meta_file_cache_throttle,
292            sst_skip_bloom_filter_in_serde: c.storage.sst_skip_bloom_filter_in_serde,
293            cache_refill_data_refill_levels: c.storage.cache_refill.data_refill_levels.clone(),
294            cache_refill_timeout_ms: c.storage.cache_refill.timeout_ms,
295            cache_refill_meta_refill_concurrency: c.storage.cache_refill.meta_refill_concurrency,
296            cache_refill_concurrency: c.storage.cache_refill.concurrency,
297            cache_refill_recent_filter_shards: c.storage.cache_refill.recent_filter_shards,
298            cache_refill_recent_filter_layers: c.storage.cache_refill.recent_filter_layers,
299            cache_refill_recent_filter_rotate_interval_ms: c
300                .storage
301                .cache_refill
302                .recent_filter_rotate_interval_ms,
303            cache_refill_unit: c.storage.cache_refill.unit,
304            cache_refill_threshold: c.storage.cache_refill.threshold,
305            cache_refill_skip_recent_filter: c.storage.cache_refill.skip_recent_filter,
306            cache_refill_skip_inheritance_filter: c.storage.cache_refill.skip_inheritance_filter,
307            cache_refill_table_cache_refill_default_policy: c
308                .streaming
309                .developer
310                .cache_refill_policy,
311            max_preload_wait_time_mill: c.storage.max_preload_wait_time_mill,
312            compact_iter_recreate_timeout_ms: c.storage.compact_iter_recreate_timeout_ms,
313
314            max_preload_io_retry_times: c.storage.max_preload_io_retry_times,
315            backup_storage_url: p.backup_storage_url().to_owned(),
316            backup_storage_directory: p.backup_storage_directory().to_owned(),
317            compactor_max_sst_key_count: c.storage.compactor_max_sst_key_count,
318            compactor_max_task_multiplier: c.storage.compactor_max_task_multiplier,
319            compactor_max_sst_size: c.storage.compactor_max_sst_size,
320            enable_fast_compaction: c.storage.enable_fast_compaction,
321            check_compaction_result: c.storage.check_compaction_result,
322            mem_table_spill_threshold: c.storage.mem_table_spill_threshold,
323            object_store_config: c.storage.object_store.clone(),
324            compactor_fast_max_compact_delete_ratio: c
325                .storage
326                .compactor_fast_max_compact_delete_ratio,
327            compactor_fast_max_compact_task_size: c.storage.compactor_fast_max_compact_task_size,
328            compactor_iter_max_io_retry_times: c.storage.compactor_iter_max_io_retry_times,
329            shorten_block_meta_key_threshold: c.storage.shorten_block_meta_key_threshold,
330            compactor_concurrent_uploading_sst_count: c
331                .storage
332                .compactor_concurrent_uploading_sst_count,
333            time_travel_version_cache_capacity: c.storage.time_travel_version_cache_capacity,
334            table_change_log_cache_capacity: c.storage.table_change_log_cache_capacity,
335            compactor_max_overlap_sst_count: c.storage.compactor_max_overlap_sst_count,
336            compactor_max_preload_meta_file_count: c.storage.compactor_max_preload_meta_file_count,
337
338            iceberg_compaction_enable_validate: c.storage.iceberg_compaction_enable_validate,
339            iceberg_compaction_max_record_batch_rows: c
340                .storage
341                .iceberg_compaction_max_record_batch_rows,
342            #[expect(deprecated)]
343            iceberg_compaction_write_parquet_max_row_group_rows: c
344                .storage
345                .iceberg_compaction_write_parquet_max_row_group_rows,
346            iceberg_compaction_min_size_per_partition_mb: c
347                .storage
348                .iceberg_compaction_min_size_per_partition_mb,
349            iceberg_compaction_max_file_count_per_partition: c
350                .storage
351                .iceberg_compaction_max_file_count_per_partition,
352            iceberg_compaction_task_parallelism_ratio: c
353                .storage
354                .iceberg_compaction_task_parallelism_ratio,
355            iceberg_compaction_enable_heuristic_output_parallelism: c
356                .storage
357                .iceberg_compaction_enable_heuristic_output_parallelism,
358            iceberg_compaction_max_concurrent_closes: c
359                .storage
360                .iceberg_compaction_max_concurrent_closes,
361            iceberg_compaction_enable_dynamic_size_estimation: c
362                .storage
363                .iceberg_compaction_enable_dynamic_size_estimation,
364            iceberg_compaction_size_estimation_smoothing_factor: c
365                .storage
366                .iceberg_compaction_size_estimation_smoothing_factor,
367            iceberg_compaction_pending_parallelism_budget_multiplier: c
368                .storage
369                .iceberg_compaction_pending_parallelism_budget_multiplier,
370            iceberg_compaction_pull_interval_ms: c.storage.iceberg_compaction_pull_interval_ms,
371            iceberg_compaction_enable_prefetch: c.storage.iceberg_compaction_enable_prefetch,
372            iceberg_compaction_target_binpack_group_size_mb: c
373                .storage
374                .iceberg_compaction_target_binpack_group_size_mb,
375            iceberg_compaction_min_group_size_mb: c.storage.iceberg_compaction_min_group_size_mb,
376            iceberg_compaction_min_group_file_count: c
377                .storage
378                .iceberg_compaction_min_group_file_count,
379            vector_file_block_size_kb: c.storage.vector_file_block_size_kb,
380            vector_block_cache_capacity_mb: s.vector_block_cache_capacity_mb,
381            vector_block_cache_shard_num: s.vector_block_cache_shard_num,
382            vector_block_cache_eviction_config: s.vector_block_cache_eviction_config.clone(),
383            vector_meta_cache_capacity_mb: s.vector_meta_cache_capacity_mb,
384            vector_meta_cache_shard_num: s.vector_meta_cache_shard_num,
385            vector_meta_cache_eviction_config: s.vector_meta_cache_eviction_config.clone(),
386        }
387    }
388}
389
390#[cfg(test)]
391mod tests {
392    use super::*;
393
394    #[test]
395    fn test_file_cache_submit_queue_size_threshold() {
396        let mut config = RwConfig::default();
397        config
398            .storage
399            .data_file_cache
400            .submit_queue_size_threshold_mb = 256;
401        config
402            .storage
403            .meta_file_cache
404            .submit_queue_size_threshold_mb = 32;
405        let system_params = system_params_for_test();
406        let storage_memory_config = extract_storage_memory_config(&config);
407
408        let opts = StorageOpts::from((&config, &system_params.into(), &storage_memory_config));
409
410        assert_eq!(opts.data_file_cache_submit_queue_size_threshold_mb, 256);
411        assert_eq!(opts.meta_file_cache_submit_queue_size_threshold_mb, 32);
412    }
413}
risingwave_storage/opts.rs

risingwave_storage/
opts.rs
