svn_cache.h revision 269847
1/** 2 * @copyright 3 * ==================================================================== 4 * Licensed to the Apache Software Foundation (ASF) under one 5 * or more contributor license agreements. See the NOTICE file 6 * distributed with this work for additional information 7 * regarding copyright ownership. The ASF licenses this file 8 * to you under the Apache License, Version 2.0 (the 9 * "License"); you may not use this file except in compliance 10 * with the License. You may obtain a copy of the License at 11 * 12 * http://www.apache.org/licenses/LICENSE-2.0 13 * 14 * Unless required by applicable law or agreed to in writing, 15 * software distributed under the License is distributed on an 16 * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY 17 * KIND, either express or implied. See the License for the 18 * specific language governing permissions and limitations 19 * under the License. 20 * ==================================================================== 21 * @endcopyright 22 * 23 * @file svn_cache.h 24 * @brief In-memory cache implementation. 25 */ 26 27 28#ifndef SVN_CACHE_H 29#define SVN_CACHE_H 30 31#include <apr_pools.h> 32#include <apr_hash.h> 33 34#include "svn_types.h" 35#include "svn_error.h" 36#include "svn_iter.h" 37#include "svn_config.h" 38#include "svn_string.h" 39 40#ifdef __cplusplus 41extern "C" { 42#endif /* __cplusplus */ 43 44 45 46/** 47 * @defgroup svn_cache__support In-memory caching 48 * @{ 49 */ 50 51/** 52 * A function type for deserializing an object @a *out from the string 53 * @a data of length @a data_len into @a result_pool. It is legal and 54 * generally suggested that the deserialization will be done in-place, 55 * i.e. modify @a data directly and return it in @a *out. 56 */ 57typedef svn_error_t *(*svn_cache__deserialize_func_t)(void **out, 58 void *data, 59 apr_size_t data_len, 60 apr_pool_t *result_pool); 61 62/** 63 * A function type for deserializing an object @a *out from the string 64 * @a data of length @a data_len into @a result_pool. The extra information 65 * @a baton passed into can be used to deserialize only a specific part or 66 * sub-structure or to perform any other non-modifying operation that may 67 * not require the whole structure to be processed. 68 */ 69typedef svn_error_t *(*svn_cache__partial_getter_func_t)(void **out, 70 const void *data, 71 apr_size_t data_len, 72 void *baton, 73 apr_pool_t *result_pool); 74 75/** 76 * A function type for modifying an already deserialized in the @a *data 77 * buffer of length @a *data_len. Additional information of the modification 78 * to do will be provided in @a baton. The function may change the size of 79 * data buffer and may re-allocate it if necessary. In that case, the new 80 * values must be passed back in @a *data_len and @a *data, respectively. 81 * Allocations will be done from @a result_pool. 82 */ 83typedef svn_error_t *(*svn_cache__partial_setter_func_t)(void **data, 84 apr_size_t *data_len, 85 void *baton, 86 apr_pool_t *result_pool); 87 88/** 89 * A function type for serializing an object @a in into bytes. The 90 * function should allocate the serialized value in @a result_pool, set 91 * @a *data to the serialized value, and set @a *data_len to its length. 92 */ 93typedef svn_error_t *(*svn_cache__serialize_func_t)(void **data, 94 apr_size_t *data_len, 95 void *in, 96 apr_pool_t *result_pool); 97 98/** 99 * A function type for transforming or ignoring errors. @a scratch_pool may 100 * be used for temporary allocations. 101 */ 102typedef svn_error_t *(*svn_cache__error_handler_t)(svn_error_t *err, 103 void *baton, 104 apr_pool_t *scratch_pool); 105 106/** 107 * A wrapper around apr_memcache_t, provided essentially so that the 108 * Subversion public API doesn't depend on whether or not you have 109 * access to the APR memcache libraries. 110 */ 111typedef struct svn_memcache_t svn_memcache_t; 112 113/** 114 * An opaque structure representing a membuffer cache object. 115 */ 116typedef struct svn_membuffer_t svn_membuffer_t; 117 118/** 119 * Opaque type for an in-memory cache. 120 */ 121typedef struct svn_cache__t svn_cache__t; 122 123/** 124 * A structure containing typical statistics about a given cache instance. 125 * Use svn_cache__get_info() to get this data. Note that not all types 126 * of caches will be able to report complete and correct information. 127 */ 128typedef struct svn_cache__info_t 129{ 130 /** A string identifying the cache instance. Usually a copy of the @a id 131 * or @a prefix parameter passed to the cache constructor. 132 */ 133 const char* id; 134 135 /** Number of getter calls (svn_cache__get() or svn_cache__get()). 136 */ 137 apr_uint64_t gets; 138 139 /** Number of getter calls that return data. 140 */ 141 apr_uint64_t hits; 142 143 /** Number of setter calls (svn_cache__set()). 144 */ 145 apr_uint64_t sets; 146 147 /** Number of function calls that returned an error. 148 */ 149 apr_uint64_t failures; 150 151 /** Size of the data currently stored in the cache. 152 * May be 0 if that information is not available. 153 */ 154 apr_uint64_t used_size; 155 156 /** Amount of memory currently reserved for cached data. 157 * Will be equal to @a used_size if no precise information is available. 158 */ 159 apr_uint64_t data_size; 160 161 /** Lower threshold of the total size of memory allocated to the cache and 162 * its index as well as management structures. The actual memory allocated 163 * by the cache may be larger. 164 */ 165 apr_uint64_t total_size; 166 167 /** Number of cache entries. 168 * May be 0 if that information is not available. 169 */ 170 apr_uint64_t used_entries; 171 172 /** Maximum numbers of cache entries. 173 * May be 0 if that information is not available. 174 */ 175 apr_uint64_t total_entries; 176} svn_cache__info_t; 177 178/** 179 * Creates a new cache in @a *cache_p. This cache will use @a pool 180 * for all of its storage needs. The elements in the cache will be 181 * indexed by keys of length @a klen, which may be APR_HASH_KEY_STRING 182 * if they are strings. Cached values will be copied in and out of 183 * the cache using @a serialize_func and @a deserialize_func, respectively. 184 * 185 * If @a deserialize_func is NULL, then the data is returned as an 186 * svn_stringbuf_t; if @a serialize_func is NULL, then the data is 187 * assumed to be an svn_stringbuf_t. 188 * 189 * The cache stores up to @a pages * @a items_per_page items at a 190 * time. The exact cache invalidation strategy is not defined here, 191 * but in general, a lower value for @a items_per_page means more 192 * memory overhead for the same number of items, but a higher value 193 * for @a items_per_page means more items are cleared at once. Both 194 * @a pages and @a items_per_page must be positive (though they both 195 * may certainly be 1). 196 * 197 * If @a thread_safe is true, and APR is compiled with threads, all 198 * accesses to the cache will be protected with a mutex. The @a id 199 * is a purely user-visible information that will allow coders to 200 * identify this cache instance in a #svn_cache__info_t struct. 201 * It does not influence the behavior of the cache itself. 202 * 203 * Note that NULL is a legitimate value for cache entries (and 204 * @a serialize_func will not be called on it). 205 * 206 * It is not safe for @a serialize_func nor @a deserialize_func to 207 * interact with the cache itself. 208 */ 209svn_error_t * 210svn_cache__create_inprocess(svn_cache__t **cache_p, 211 svn_cache__serialize_func_t serialize_func, 212 svn_cache__deserialize_func_t deserialize_func, 213 apr_ssize_t klen, 214 apr_int64_t pages, 215 apr_int64_t items_per_page, 216 svn_boolean_t thread_safe, 217 const char *id, 218 apr_pool_t *pool); 219 220/** 221 * Creates a new cache in @a *cache_p, communicating to a memcached 222 * process via @a memcache. The elements in the cache will be indexed 223 * by keys of length @a klen, which may be APR_HASH_KEY_STRING if they 224 * are strings. Values will be serialized for memcached using @a 225 * serialize_func and deserialized using @a deserialize_func. Because 226 * the same memcached server may cache many different kinds of values, 227 * @a prefix should be specified to differentiate this cache from 228 * other caches. @a *cache_p will be allocated in @a result_pool. 229 * 230 * If @a deserialize_func is NULL, then the data is returned as an 231 * svn_stringbuf_t; if @a serialize_func is NULL, then the data is 232 * assumed to be an svn_stringbuf_t. 233 * 234 * These caches are always thread safe. 235 * 236 * These caches do not support svn_cache__iter. 237 * 238 * If Subversion was not built with apr_memcache support, always 239 * raises SVN_ERR_NO_APR_MEMCACHE. 240 */ 241svn_error_t * 242svn_cache__create_memcache(svn_cache__t **cache_p, 243 svn_memcache_t *memcache, 244 svn_cache__serialize_func_t serialize_func, 245 svn_cache__deserialize_func_t deserialize_func, 246 apr_ssize_t klen, 247 const char *prefix, 248 apr_pool_t *result_pool); 249 250/** 251 * Given @a config, returns an APR memcached interface in @a 252 * *memcache_p allocated in @a result_pool if @a config contains entries in 253 * the SVN_CACHE_CONFIG_CATEGORY_MEMCACHED_SERVERS section describing 254 * memcached servers; otherwise, sets @a *memcache_p to NULL. 255 * 256 * If Subversion was not built with apr_memcache_support, then raises 257 * SVN_ERR_NO_APR_MEMCACHE if and only if @a config is configured to 258 * use memcache. 259 */ 260svn_error_t * 261svn_cache__make_memcache_from_config(svn_memcache_t **memcache_p, 262 svn_config_t *config, 263 apr_pool_t *result_pool); 264 265/** 266 * Creates a new membuffer cache object in @a *cache. It will contain 267 * up to @a total_size bytes of data, using @a directory_size bytes 268 * for index information and the remainder for serialized objects. 269 * 270 * Since each index entry is about 50 bytes long, 1 to 10 percent of 271 * the @a total_size should be allocated to the @a directory_size, 272 * depending on the average serialized object size. Higher percentages 273 * will generally result in higher hit rates and reduced conflict 274 * resolution overhead. 275 * 276 * The cache will be split into @a segment_count segments of equal size. 277 * A higher number reduces lock contention but also limits the maximum 278 * cachable item size. If it is not a power of two, it will be rounded 279 * down to next lower power of two. Also, there is an implementation 280 * specific upper limit and the setting will be capped there automatically. 281 * If the number is 0, a default will be derived from @a total_size. 282 * 283 * If access to the resulting cache object is guaranteed to be serialized, 284 * @a thread_safe may be set to @c FALSE for maximum performance. 285 * 286 * There is no limit on the number of threads reading a given cache segment 287 * concurrently. Writes, however, need an exclusive lock on the respective 288 * segment. @a allow_blocking_writes controls contention is handled here. 289 * If set to TRUE, writes will wait until the lock becomes available, i.e. 290 * reads should be short. If set to FALSE, write attempts will be ignored 291 * (no data being written to the cache) if some reader or another writer 292 * currently holds the segment lock. 293 * 294 * Allocations will be made in @a result_pool, in particular the data buffers. 295 */ 296svn_error_t * 297svn_cache__membuffer_cache_create(svn_membuffer_t **cache, 298 apr_size_t total_size, 299 apr_size_t directory_size, 300 apr_size_t segment_count, 301 svn_boolean_t thread_safe, 302 svn_boolean_t allow_blocking_writes, 303 apr_pool_t *result_pool); 304 305/** 306 * Creates a new cache in @a *cache_p, storing the data in a potentially 307 * shared @a membuffer object. The elements in the cache will be indexed 308 * by keys of length @a klen, which may be APR_HASH_KEY_STRING if they 309 * are strings. Values will be serialized for the memcache using @a 310 * serialize_func and deserialized using @a deserialize_func. Because 311 * the same memcache object may cache many different kinds of values 312 * form multiple caches, @a prefix should be specified to differentiate 313 * this cache from other caches. @a *cache_p will be allocated in @a result_pool. 314 * 315 * If @a deserialize_func is NULL, then the data is returned as an 316 * svn_stringbuf_t; if @a serialize_func is NULL, then the data is 317 * assumed to be an svn_stringbuf_t. 318 * 319 * If @a thread_safe is true, and APR is compiled with threads, all 320 * accesses to the cache will be protected with a mutex, if the shared 321 * @a memcache has also been created with thread_safe flag set. 322 * 323 * These caches do not support svn_cache__iter. 324 */ 325svn_error_t * 326svn_cache__create_membuffer_cache(svn_cache__t **cache_p, 327 svn_membuffer_t *membuffer, 328 svn_cache__serialize_func_t serialize, 329 svn_cache__deserialize_func_t deserialize, 330 apr_ssize_t klen, 331 const char *prefix, 332 svn_boolean_t thread_safe, 333 apr_pool_t *result_pool); 334 335/** 336 * Sets @a handler to be @a cache's error handling routine. If any 337 * error is returned from a call to svn_cache__get or svn_cache__set, @a 338 * handler will be called with @a baton and the error, and the 339 * original function will return whatever error @a handler returns 340 * instead (possibly SVN_NO_ERROR); @a handler will receive the pool 341 * passed to the svn_cache_* function. @a scratch_pool is used for temporary 342 * allocations. 343 */ 344svn_error_t * 345svn_cache__set_error_handler(svn_cache__t *cache, 346 svn_cache__error_handler_t handler, 347 void *baton, 348 apr_pool_t *scratch_pool); 349 350/** 351 * Returns @c TRUE if the @a cache supports objects of the given @a size. 352 * There is no guarantee, that svn_cache__set() will actually store the 353 * respective object in that case. However, a @c FALSE return value indicates 354 * that an attempt to cache the item will either fail or impair the overall 355 * cache performance. @c FALSE will also be returned if @a cache is @c NULL. 356 */ 357svn_boolean_t 358svn_cache__is_cachable(svn_cache__t *cache, 359 apr_size_t size); 360 361#define SVN_CACHE_CONFIG_CATEGORY_MEMCACHED_SERVERS "memcached-servers" 362 363/** 364 * Fetches a value indexed by @a key from @a cache into @a *value, 365 * setting @a *found to TRUE iff it is in the cache and FALSE if it is 366 * not found. @a key may be NULL in which case @a *found will be 367 * FALSE. The value is copied into @a result_pool using the deserialize 368 * function provided to the cache's constructor. 369 */ 370svn_error_t * 371svn_cache__get(void **value, 372 svn_boolean_t *found, 373 svn_cache__t *cache, 374 const void *key, 375 apr_pool_t *result_pool); 376 377/** 378 * Stores the value @a value under the key @a key in @a cache. Uses @a 379 * scratch_pool for temporary allocations. The cache makes copies of 380 * @a key and @a value if necessary (that is, @a key and @a value may 381 * have shorter lifetimes than the cache). @a key may be NULL in which 382 * case the cache will remain unchanged. 383 * 384 * If there is already a value for @a key, this will replace it. Bear 385 * in mind that in some circumstances this may leak memory (that is, 386 * the cache's copy of the previous value may not be immediately 387 * cleared); it is only guaranteed to not leak for caches created with 388 * @a items_per_page equal to 1. 389 */ 390svn_error_t * 391svn_cache__set(svn_cache__t *cache, 392 const void *key, 393 void *value, 394 apr_pool_t *scratch_pool); 395 396/** 397 * Iterates over the elements currently in @a cache, calling @a func 398 * for each one until there are no more elements or @a func returns an 399 * error. Uses @a scratch_pool for temporary allocations. 400 * 401 * If @a completed is not NULL, then on return - if @a func returns no 402 * errors - @a *completed will be set to @c TRUE. 403 * 404 * If @a func returns an error other than @c SVN_ERR_ITER_BREAK, that 405 * error is returned. When @a func returns @c SVN_ERR_ITER_BREAK, 406 * iteration is interrupted, but no error is returned and @a 407 * *completed is set to @c FALSE. (The error handler set by 408 * svn_cache__set_error_handler is not used for svn_cache__iter.) 409 * 410 * It is not legal to perform any other cache operations on @a cache 411 * inside @a func. 412 * 413 * svn_cache__iter is not supported by all cache implementations; see 414 * the svn_cache__create_* function for details. 415 */ 416svn_error_t * 417svn_cache__iter(svn_boolean_t *completed, 418 svn_cache__t *cache, 419 svn_iter_apr_hash_cb_t func, 420 void *baton, 421 apr_pool_t *scratch_pool); 422 423/** 424 * Similar to svn_cache__get() but will call a specific de-serialization 425 * function @a func. @a found will be set depending on whether the @a key 426 * has been found. Even if that reports @c TRUE, @a value may still return 427 * a @c NULL pointer depending on the logic inside @a func. For a @a NULL 428 * @a key, no data will be found. @a value will be allocated in 429 * @a result_pool. 430 */ 431svn_error_t * 432svn_cache__get_partial(void **value, 433 svn_boolean_t *found, 434 svn_cache__t *cache, 435 const void *key, 436 svn_cache__partial_getter_func_t func, 437 void *baton, 438 apr_pool_t *result_pool); 439 440/** 441 * Find the item identified by @a key in the @a cache. If it has been found, 442 * call @a func for it and @a baton to potentially modify the data. Changed 443 * data will be written back to the cache. If the item cannot be found, 444 * or if @a key is NULL, @a func does not get called. @a scratch_pool is 445 * used for temporary allocations. 446 */ 447svn_error_t * 448svn_cache__set_partial(svn_cache__t *cache, 449 const void *key, 450 svn_cache__partial_setter_func_t func, 451 void *baton, 452 apr_pool_t *scratch_pool); 453 454/** 455 * Collect all available usage statistics on the cache instance @a cache 456 * and write the data into @a info. If @a reset has been set, access 457 * counters will be reset right after copying the statistics info. 458 * @a result_pool will be used for allocations. 459 */ 460svn_error_t * 461svn_cache__get_info(svn_cache__t *cache, 462 svn_cache__info_t *info, 463 svn_boolean_t reset, 464 apr_pool_t *result_pool); 465 466/** 467 * Return the information given in @a info formatted as a multi-line string. 468 * Allocations take place in @a result_pool. 469 */ 470svn_string_t * 471svn_cache__format_info(const svn_cache__info_t *info, 472 apr_pool_t *result_pool); 473 474/* Access the process-global (singleton) membuffer cache. The first call 475 * will automatically allocate the cache using the current cache config. 476 * NULL will be returned if the desired cache size is 0. 477 * 478 * @since New in 1.7. 479 */ 480struct svn_membuffer_t * 481svn_cache__get_global_membuffer_cache(void); 482 483/** @} */ 484 485 486#ifdef __cplusplus 487} 488#endif /* __cplusplus */ 489 490#endif /* SVN_CACHE_H */ 491