| 1 | /* |
| 2 | * librdkafka - Apache Kafka C library |
| 3 | * |
| 4 | * Copyright (c) 2012-2013, Magnus Edenhill |
| 5 | * All rights reserved. |
| 6 | * |
| 7 | * Redistribution and use in source and binary forms, with or without |
| 8 | * modification, are permitted provided that the following conditions are met: |
| 9 | * |
| 10 | * 1. Redistributions of source code must retain the above copyright notice, |
| 11 | * this list of conditions and the following disclaimer. |
| 12 | * 2. Redistributions in binary form must reproduce the above copyright notice, |
| 13 | * this list of conditions and the following disclaimer in the documentation |
| 14 | * and/or other materials provided with the distribution. |
| 15 | * |
| 16 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 17 | * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 18 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 19 | * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE |
| 20 | * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 21 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 22 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| 23 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| 24 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 25 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
| 26 | * POSSIBILITY OF SUCH DAMAGE. |
| 27 | */ |
| 28 | |
| 29 | |
| 30 | #include "rd.h" |
| 31 | #include "rdkafka_int.h" |
| 32 | #include "rdkafka_topic.h" |
| 33 | #include "rdkafka_broker.h" |
| 34 | #include "rdkafka_request.h" |
| 35 | #include "rdkafka_metadata.h" |
| 36 | |
| 37 | #include <string.h> |
| 38 | /** |
| 39 | * @{ |
| 40 | * |
| 41 | * @brief Metadata cache |
| 42 | * |
| 43 | * The metadata cache consists of cached topic metadata as |
| 44 | * retrieved from the cluster using MetadataRequest. |
| 45 | * |
| 46 | * The topic cache entries are made up \c struct rd_kafka_metadata_cache_entry |
| 47 | * each containing the topic name, a copy of the topic's metadata |
| 48 | * and a cache expiry time. |
| 49 | * |
| 50 | * On update any previous entry for the topic are removed and replaced |
| 51 | * with a new entry. |
| 52 | * |
| 53 | * The cache is also populated when the topic metadata is being requested |
| 54 | * for specific topics, this will not interfere with existing cache entries |
| 55 | * for topics, but for any topics not currently in the cache a new |
| 56 | * entry will be added with a flag (RD_KAFKA_METADATA_CACHE_VALID(rkmce)) |
| 57 | * indicating that the entry is waiting to be populated by the MetadataResponse. |
| 58 | * |
| 59 | * The cache is locked in its entirety with rd_kafka_wr/rdlock() by the caller |
| 60 | * and the returned cache entry must only be accessed during the duration |
| 61 | * of the lock. |
| 62 | * |
| 63 | */ |
| 64 | |
| 65 | static void rd_kafka_metadata_cache_propagate_changes (rd_kafka_t *rk); |
| 66 | |
| 67 | |
| 68 | /** |
| 69 | * @brief Remove and free cache entry. |
| 70 | * |
| 71 | * @remark The expiry timer is not updated, for simplicity. |
| 72 | * @locks rd_kafka_wrlock() |
| 73 | */ |
| 74 | static RD_INLINE void |
| 75 | rd_kafka_metadata_cache_delete (rd_kafka_t *rk, |
| 76 | struct rd_kafka_metadata_cache_entry *rkmce, |
| 77 | int unlink_avl) { |
| 78 | if (unlink_avl) |
| 79 | RD_AVL_REMOVE_ELM(&rk->rk_metadata_cache.rkmc_avl, rkmce); |
| 80 | TAILQ_REMOVE(&rk->rk_metadata_cache.rkmc_expiry, rkmce, rkmce_link); |
| 81 | rd_kafka_assert(NULL, rk->rk_metadata_cache.rkmc_cnt > 0); |
| 82 | rk->rk_metadata_cache.rkmc_cnt--; |
| 83 | |
| 84 | rd_free(rkmce); |
| 85 | } |
| 86 | |
| 87 | /** |
| 88 | * @brief Delete cache entry by topic name |
| 89 | * @locks rd_kafka_wrlock() |
| 90 | * @returns 1 if entry was found and removed, else 0. |
| 91 | */ |
| 92 | static int rd_kafka_metadata_cache_delete_by_name (rd_kafka_t *rk, |
| 93 | const char *topic) { |
| 94 | struct rd_kafka_metadata_cache_entry *rkmce; |
| 95 | |
| 96 | rkmce = rd_kafka_metadata_cache_find(rk, topic, 1); |
| 97 | if (rkmce) |
| 98 | rd_kafka_metadata_cache_delete(rk, rkmce, 1); |
| 99 | return rkmce ? 1 : 0; |
| 100 | } |
| 101 | |
| 102 | static int rd_kafka_metadata_cache_evict (rd_kafka_t *rk); |
| 103 | |
| 104 | /** |
| 105 | * @brief Cache eviction timer callback. |
| 106 | * @locality rdkafka main thread |
| 107 | * @locks NOT rd_kafka_*lock() |
| 108 | */ |
| 109 | static void rd_kafka_metadata_cache_evict_tmr_cb (rd_kafka_timers_t *rkts, |
| 110 | void *arg) { |
| 111 | rd_kafka_t *rk = arg; |
| 112 | |
| 113 | rd_kafka_wrlock(rk); |
| 114 | rd_kafka_metadata_cache_evict(rk); |
| 115 | rd_kafka_wrunlock(rk); |
| 116 | } |
| 117 | |
| 118 | |
| 119 | /** |
| 120 | * @brief Evict timed out entries from cache and rearm timer for |
| 121 | * next expiry. |
| 122 | * |
| 123 | * @returns the number of entries evicted. |
| 124 | * |
| 125 | * @locks rd_kafka_wrlock() |
| 126 | */ |
| 127 | static int rd_kafka_metadata_cache_evict (rd_kafka_t *rk) { |
| 128 | int cnt = 0; |
| 129 | rd_ts_t now = rd_clock(); |
| 130 | struct rd_kafka_metadata_cache_entry *rkmce; |
| 131 | |
| 132 | while ((rkmce = TAILQ_FIRST(&rk->rk_metadata_cache.rkmc_expiry)) && |
| 133 | rkmce->rkmce_ts_expires <= now) { |
| 134 | rd_kafka_metadata_cache_delete(rk, rkmce, 1); |
| 135 | cnt++; |
| 136 | } |
| 137 | |
| 138 | if (rkmce) |
| 139 | rd_kafka_timer_start(&rk->rk_timers, |
| 140 | &rk->rk_metadata_cache.rkmc_expiry_tmr, |
| 141 | rkmce->rkmce_ts_expires - now, |
| 142 | rd_kafka_metadata_cache_evict_tmr_cb, |
| 143 | rk); |
| 144 | else |
| 145 | rd_kafka_timer_stop(&rk->rk_timers, |
| 146 | &rk->rk_metadata_cache.rkmc_expiry_tmr, 1); |
| 147 | |
| 148 | rd_kafka_dbg(rk, METADATA, "METADATA" , |
| 149 | "Expired %d entries from metadata cache " |
| 150 | "(%d entries remain)" , |
| 151 | cnt, rk->rk_metadata_cache.rkmc_cnt); |
| 152 | |
| 153 | if (cnt) |
| 154 | rd_kafka_metadata_cache_propagate_changes(rk); |
| 155 | |
| 156 | return cnt; |
| 157 | } |
| 158 | |
| 159 | |
| 160 | /** |
| 161 | * @brief Find cache entry by topic name |
| 162 | * |
| 163 | * @param valid: entry must be valid (not hint) |
| 164 | * |
| 165 | * @locks rd_kafka_*lock() |
| 166 | */ |
| 167 | struct rd_kafka_metadata_cache_entry * |
| 168 | rd_kafka_metadata_cache_find (rd_kafka_t *rk, const char *topic, int valid) { |
| 169 | struct rd_kafka_metadata_cache_entry skel, *rkmce; |
| 170 | skel.rkmce_mtopic.topic = (char *)topic; |
| 171 | rkmce = RD_AVL_FIND(&rk->rk_metadata_cache.rkmc_avl, &skel); |
| 172 | if (rkmce && (!valid || RD_KAFKA_METADATA_CACHE_VALID(rkmce))) |
| 173 | return rkmce; |
| 174 | return NULL; |
| 175 | } |
| 176 | |
| 177 | |
| 178 | /** |
| 179 | * @brief Partition (id) comparator |
| 180 | */ |
| 181 | int rd_kafka_metadata_partition_id_cmp (const void *_a, |
| 182 | const void *_b) { |
| 183 | const rd_kafka_metadata_partition_t *a = _a, *b = _b; |
| 184 | return a->id - b->id; |
| 185 | } |
| 186 | |
| 187 | |
| 188 | /** |
| 189 | * @brief Add (and replace) cache entry for topic. |
| 190 | * |
| 191 | * This makes a copy of \p topic |
| 192 | * |
| 193 | * @locks rd_kafka_wrlock() |
| 194 | */ |
| 195 | static struct rd_kafka_metadata_cache_entry * |
| 196 | rd_kafka_metadata_cache_insert (rd_kafka_t *rk, |
| 197 | const rd_kafka_metadata_topic_t *mtopic, |
| 198 | rd_ts_t now, rd_ts_t ts_expires) { |
| 199 | struct rd_kafka_metadata_cache_entry *rkmce, *old; |
| 200 | size_t topic_len; |
| 201 | rd_tmpabuf_t tbuf; |
| 202 | int i; |
| 203 | |
| 204 | /* Metadata is stored in one contigious buffer where structs and |
| 205 | * and pointed-to fields are layed out in a memory aligned fashion. |
| 206 | * rd_tmpabuf_t provides the infrastructure to do this. |
| 207 | * Because of this we copy all the structs verbatim but |
| 208 | * any pointer fields needs to be copied explicitly to update |
| 209 | * the pointer address. */ |
| 210 | topic_len = strlen(mtopic->topic) + 1; |
| 211 | rd_tmpabuf_new(&tbuf, |
| 212 | RD_ROUNDUP(sizeof(*rkmce), 8) + |
| 213 | RD_ROUNDUP(topic_len, 8) + |
| 214 | (mtopic->partition_cnt * |
| 215 | RD_ROUNDUP(sizeof(*mtopic->partitions), 8)), |
| 216 | 1/*assert on fail*/); |
| 217 | |
| 218 | rkmce = rd_tmpabuf_alloc(&tbuf, sizeof(*rkmce)); |
| 219 | |
| 220 | rkmce->rkmce_mtopic = *mtopic; |
| 221 | |
| 222 | /* Copy topic name and update pointer */ |
| 223 | rkmce->rkmce_mtopic.topic = rd_tmpabuf_write_str(&tbuf, mtopic->topic); |
| 224 | |
| 225 | /* Copy partition array and update pointer */ |
| 226 | rkmce->rkmce_mtopic.partitions = |
| 227 | rd_tmpabuf_write(&tbuf, mtopic->partitions, |
| 228 | mtopic->partition_cnt * |
| 229 | sizeof(*mtopic->partitions)); |
| 230 | |
| 231 | /* Clear uncached fields. */ |
| 232 | for (i = 0 ; i < mtopic->partition_cnt ; i++) { |
| 233 | rkmce->rkmce_mtopic.partitions[i].replicas = NULL; |
| 234 | rkmce->rkmce_mtopic.partitions[i].replica_cnt = 0; |
| 235 | rkmce->rkmce_mtopic.partitions[i].isrs = NULL; |
| 236 | rkmce->rkmce_mtopic.partitions[i].isr_cnt = 0; |
| 237 | } |
| 238 | |
| 239 | /* Sort partitions for future bsearch() lookups. */ |
| 240 | qsort(rkmce->rkmce_mtopic.partitions, |
| 241 | rkmce->rkmce_mtopic.partition_cnt, |
| 242 | sizeof(*rkmce->rkmce_mtopic.partitions), |
| 243 | rd_kafka_metadata_partition_id_cmp); |
| 244 | |
| 245 | TAILQ_INSERT_TAIL(&rk->rk_metadata_cache.rkmc_expiry, |
| 246 | rkmce, rkmce_link); |
| 247 | rk->rk_metadata_cache.rkmc_cnt++; |
| 248 | rkmce->rkmce_ts_expires = ts_expires; |
| 249 | rkmce->rkmce_ts_insert = now; |
| 250 | |
| 251 | /* Insert (and replace existing) entry. */ |
| 252 | old = RD_AVL_INSERT(&rk->rk_metadata_cache.rkmc_avl, rkmce, |
| 253 | rkmce_avlnode); |
| 254 | if (old) |
| 255 | rd_kafka_metadata_cache_delete(rk, old, 0); |
| 256 | |
| 257 | /* Explicitly not freeing the tmpabuf since rkmce points to its |
| 258 | * memory. */ |
| 259 | return rkmce; |
| 260 | } |
| 261 | |
| 262 | |
| 263 | /** |
| 264 | * @brief Purge the metadata cache |
| 265 | * |
| 266 | * @locks rd_kafka_wrlock() |
| 267 | */ |
| 268 | static void rd_kafka_metadata_cache_purge (rd_kafka_t *rk) { |
| 269 | struct rd_kafka_metadata_cache_entry *rkmce; |
| 270 | int was_empty = TAILQ_EMPTY(&rk->rk_metadata_cache.rkmc_expiry); |
| 271 | |
| 272 | while ((rkmce = TAILQ_FIRST(&rk->rk_metadata_cache.rkmc_expiry))) |
| 273 | rd_kafka_metadata_cache_delete(rk, rkmce, 1); |
| 274 | |
| 275 | rd_kafka_timer_stop(&rk->rk_timers, |
| 276 | &rk->rk_metadata_cache.rkmc_expiry_tmr, 1); |
| 277 | |
| 278 | if (!was_empty) |
| 279 | rd_kafka_metadata_cache_propagate_changes(rk); |
| 280 | } |
| 281 | |
| 282 | |
| 283 | /** |
| 284 | * @brief Start or update the cache expiry timer. |
| 285 | * Typically done after a series of cache_topic_update() |
| 286 | * |
| 287 | * @locks rd_kafka_wrlock() |
| 288 | */ |
| 289 | void rd_kafka_metadata_cache_expiry_start (rd_kafka_t *rk) { |
| 290 | struct rd_kafka_metadata_cache_entry *rkmce; |
| 291 | |
| 292 | if ((rkmce = TAILQ_FIRST(&rk->rk_metadata_cache.rkmc_expiry))) |
| 293 | rd_kafka_timer_start(&rk->rk_timers, |
| 294 | &rk->rk_metadata_cache.rkmc_expiry_tmr, |
| 295 | rkmce->rkmce_ts_expires - rd_clock(), |
| 296 | rd_kafka_metadata_cache_evict_tmr_cb, |
| 297 | rk); |
| 298 | } |
| 299 | |
| 300 | /** |
| 301 | * @brief Update the metadata cache for a single topic |
| 302 | * with the provided metadata. |
| 303 | * If the topic has an error the existing entry is removed |
| 304 | * and no new entry is added, which avoids the topic to be |
| 305 | * suppressed in upcoming metadata requests because being in the cache. |
| 306 | * In other words: we want to re-query errored topics. |
| 307 | * |
| 308 | * @remark The cache expiry timer will not be updated/started, |
| 309 | * call rd_kafka_metadata_cache_expiry_start() instead. |
| 310 | * |
| 311 | * @locks rd_kafka_wrlock() |
| 312 | */ |
| 313 | void |
| 314 | rd_kafka_metadata_cache_topic_update (rd_kafka_t *rk, |
| 315 | const rd_kafka_metadata_topic_t *mdt) { |
| 316 | rd_ts_t now = rd_clock(); |
| 317 | rd_ts_t ts_expires = now + (rk->rk_conf.metadata_max_age_ms * 1000); |
| 318 | int changed = 1; |
| 319 | |
| 320 | if (!mdt->err) |
| 321 | rd_kafka_metadata_cache_insert(rk, mdt, now, ts_expires); |
| 322 | else |
| 323 | changed = rd_kafka_metadata_cache_delete_by_name(rk, |
| 324 | mdt->topic); |
| 325 | |
| 326 | if (changed) |
| 327 | rd_kafka_metadata_cache_propagate_changes(rk); |
| 328 | } |
| 329 | |
| 330 | |
| 331 | /** |
| 332 | * @brief Update the metadata cache with the provided metadata. |
| 333 | * |
| 334 | * @param abs_update int: absolute update: purge cache before updating. |
| 335 | * |
| 336 | * @locks rd_kafka_wrlock() |
| 337 | */ |
| 338 | void rd_kafka_metadata_cache_update (rd_kafka_t *rk, |
| 339 | const rd_kafka_metadata_t *md, |
| 340 | int abs_update) { |
| 341 | struct rd_kafka_metadata_cache_entry *rkmce; |
| 342 | rd_ts_t now = rd_clock(); |
| 343 | rd_ts_t ts_expires = now + (rk->rk_conf.metadata_max_age_ms * 1000); |
| 344 | int i; |
| 345 | |
| 346 | rd_kafka_dbg(rk, METADATA, "METADATA" , |
| 347 | "%s of metadata cache with %d topic(s)" , |
| 348 | abs_update ? "Absolute update" : "Update" , |
| 349 | md->topic_cnt); |
| 350 | |
| 351 | if (abs_update) |
| 352 | rd_kafka_metadata_cache_purge(rk); |
| 353 | |
| 354 | |
| 355 | for (i = 0 ; i < md->topic_cnt ; i++) |
| 356 | rd_kafka_metadata_cache_insert(rk, &md->topics[i], now, |
| 357 | ts_expires); |
| 358 | |
| 359 | /* Update expiry timer */ |
| 360 | if ((rkmce = TAILQ_FIRST(&rk->rk_metadata_cache.rkmc_expiry))) |
| 361 | rd_kafka_timer_start(&rk->rk_timers, |
| 362 | &rk->rk_metadata_cache.rkmc_expiry_tmr, |
| 363 | rkmce->rkmce_ts_expires - now, |
| 364 | rd_kafka_metadata_cache_evict_tmr_cb, |
| 365 | rk); |
| 366 | |
| 367 | if (md->topic_cnt > 0) |
| 368 | rd_kafka_metadata_cache_propagate_changes(rk); |
| 369 | } |
| 370 | |
| 371 | |
| 372 | /** |
| 373 | * @brief Remove cache hints for topics in \p topics |
| 374 | * This is done when the Metadata response has been parsed and |
| 375 | * replaced hints with existing topic information, thus this will |
| 376 | * only remove unmatched topics from the cache. |
| 377 | * |
| 378 | * @locks rd_kafka_wrlock() |
| 379 | */ |
| 380 | void rd_kafka_metadata_cache_purge_hints (rd_kafka_t *rk, |
| 381 | const rd_list_t *topics) { |
| 382 | const char *topic; |
| 383 | int i; |
| 384 | int cnt = 0; |
| 385 | |
| 386 | RD_LIST_FOREACH(topic, topics, i) { |
| 387 | struct rd_kafka_metadata_cache_entry *rkmce; |
| 388 | |
| 389 | if (!(rkmce = rd_kafka_metadata_cache_find(rk, topic, |
| 390 | 0/*any*/)) || |
| 391 | RD_KAFKA_METADATA_CACHE_VALID(rkmce)) |
| 392 | continue; |
| 393 | |
| 394 | rd_kafka_metadata_cache_delete(rk, rkmce, 1/*unlink avl*/); |
| 395 | cnt++; |
| 396 | } |
| 397 | |
| 398 | if (cnt > 0) { |
| 399 | rd_kafka_dbg(rk, METADATA, "METADATA" , |
| 400 | "Purged %d/%d cached topic hint(s)" , |
| 401 | cnt, rd_list_cnt(topics)); |
| 402 | rd_kafka_metadata_cache_propagate_changes(rk); |
| 403 | } |
| 404 | } |
| 405 | |
| 406 | |
| 407 | /** |
| 408 | * @brief Inserts a non-valid entry for topics in \p topics indicating |
| 409 | * that a MetadataRequest is in progress. |
| 410 | * This avoids sending multiple MetadataRequests for the same topics |
| 411 | * if there are already outstanding requests, see |
| 412 | * \c rd_kafka_metadata_refresh_topics(). |
| 413 | * |
| 414 | * @remark These non-valid cache entries' expire time is set to the |
| 415 | * MetadataRequest timeout. |
| 416 | * |
| 417 | * @param dst rd_list_t(char *topicname): if not NULL: populated with |
| 418 | * topics that were added as hints to cache, e.q., topics to query. |
| 419 | * @param topics rd_list_t(char *topicname) |
| 420 | * @param replace int: replace existing valid entries |
| 421 | * |
| 422 | * @returns the number of topic hints inserted. |
| 423 | * |
| 424 | * @locks rd_kafka_wrlock() |
| 425 | */ |
| 426 | int rd_kafka_metadata_cache_hint (rd_kafka_t *rk, |
| 427 | const rd_list_t *topics, rd_list_t *dst, |
| 428 | int replace) { |
| 429 | const char *topic; |
| 430 | rd_ts_t now = rd_clock(); |
| 431 | rd_ts_t ts_expires = now + (rk->rk_conf.socket_timeout_ms * 1000); |
| 432 | int i; |
| 433 | int cnt = 0; |
| 434 | |
| 435 | RD_LIST_FOREACH(topic, topics, i) { |
| 436 | rd_kafka_metadata_topic_t mtopic = { |
| 437 | .topic = (char *)topic, |
| 438 | .err = RD_KAFKA_RESP_ERR__WAIT_CACHE |
| 439 | }; |
| 440 | const struct rd_kafka_metadata_cache_entry *rkmce; |
| 441 | |
| 442 | /* !replace: Dont overwrite valid entries */ |
| 443 | if (!replace && |
| 444 | (rkmce = |
| 445 | rd_kafka_metadata_cache_find(rk, topic, 0/*any*/))) { |
| 446 | if (RD_KAFKA_METADATA_CACHE_VALID(rkmce) || dst) |
| 447 | continue; |
| 448 | /* FALLTHRU */ |
| 449 | } |
| 450 | |
| 451 | rd_kafka_metadata_cache_insert(rk, &mtopic, now, ts_expires); |
| 452 | cnt++; |
| 453 | |
| 454 | if (dst) |
| 455 | rd_list_add(dst, rd_strdup(topic)); |
| 456 | |
| 457 | } |
| 458 | |
| 459 | if (cnt > 0) |
| 460 | rd_kafka_dbg(rk, METADATA, "METADATA" , |
| 461 | "Hinted cache of %d/%d topic(s) being queried" , |
| 462 | cnt, rd_list_cnt(topics)); |
| 463 | |
| 464 | return cnt; |
| 465 | } |
| 466 | |
| 467 | |
| 468 | /** |
| 469 | * @brief Same as rd_kafka_metadata_cache_hint() but takes |
| 470 | * a topic+partition list as input instead. |
| 471 | */ |
| 472 | int rd_kafka_metadata_cache_hint_rktparlist ( |
| 473 | rd_kafka_t *rk, |
| 474 | const rd_kafka_topic_partition_list_t *rktparlist, |
| 475 | rd_list_t *dst, |
| 476 | int replace) { |
| 477 | rd_list_t topics; |
| 478 | int r; |
| 479 | |
| 480 | rd_list_init(&topics, rktparlist->cnt, rd_free); |
| 481 | rd_kafka_topic_partition_list_get_topic_names(rktparlist, &topics, |
| 482 | 0/*dont include regex*/); |
| 483 | r = rd_kafka_metadata_cache_hint(rk, &topics, dst, replace); |
| 484 | rd_list_destroy(&topics); |
| 485 | return r; |
| 486 | } |
| 487 | |
| 488 | |
| 489 | /** |
| 490 | * @brief Cache entry comparator (on topic name) |
| 491 | */ |
| 492 | static int rd_kafka_metadata_cache_entry_cmp (const void *_a, const void *_b) { |
| 493 | const struct rd_kafka_metadata_cache_entry *a = _a, *b = _b; |
| 494 | return strcmp(a->rkmce_mtopic.topic, b->rkmce_mtopic.topic); |
| 495 | } |
| 496 | |
| 497 | |
| 498 | /** |
| 499 | * @brief Initialize the metadata cache |
| 500 | * |
| 501 | * @locks rd_kafka_wrlock() |
| 502 | */ |
| 503 | void rd_kafka_metadata_cache_init (rd_kafka_t *rk) { |
| 504 | rd_avl_init(&rk->rk_metadata_cache.rkmc_avl, |
| 505 | rd_kafka_metadata_cache_entry_cmp, 0); |
| 506 | TAILQ_INIT(&rk->rk_metadata_cache.rkmc_expiry); |
| 507 | mtx_init(&rk->rk_metadata_cache.rkmc_full_lock, mtx_plain); |
| 508 | mtx_init(&rk->rk_metadata_cache.rkmc_cnd_lock, mtx_plain); |
| 509 | cnd_init(&rk->rk_metadata_cache.rkmc_cnd); |
| 510 | |
| 511 | } |
| 512 | |
| 513 | /** |
| 514 | * @brief Purge and destroy metadata cache |
| 515 | * |
| 516 | * @locks rd_kafka_wrlock() |
| 517 | */ |
| 518 | void rd_kafka_metadata_cache_destroy (rd_kafka_t *rk) { |
| 519 | rd_kafka_timer_stop(&rk->rk_timers, |
| 520 | &rk->rk_metadata_cache.rkmc_query_tmr, 1/*lock*/); |
| 521 | rd_kafka_metadata_cache_purge(rk); |
| 522 | mtx_destroy(&rk->rk_metadata_cache.rkmc_full_lock); |
| 523 | mtx_destroy(&rk->rk_metadata_cache.rkmc_cnd_lock); |
| 524 | cnd_destroy(&rk->rk_metadata_cache.rkmc_cnd); |
| 525 | rd_avl_destroy(&rk->rk_metadata_cache.rkmc_avl); |
| 526 | } |
| 527 | |
| 528 | |
| 529 | /** |
| 530 | * @brief Wait for cache update, or timeout. |
| 531 | * |
| 532 | * @returns 1 on cache update or 0 on timeout. |
| 533 | * @locks none |
| 534 | * @locality any |
| 535 | */ |
| 536 | int rd_kafka_metadata_cache_wait_change (rd_kafka_t *rk, int timeout_ms) { |
| 537 | int r; |
| 538 | #if ENABLE_DEVEL |
| 539 | rd_ts_t ts_start = rd_clock(); |
| 540 | #endif |
| 541 | mtx_lock(&rk->rk_metadata_cache.rkmc_cnd_lock); |
| 542 | r = cnd_timedwait_ms(&rk->rk_metadata_cache.rkmc_cnd, |
| 543 | &rk->rk_metadata_cache.rkmc_cnd_lock, |
| 544 | timeout_ms); |
| 545 | mtx_unlock(&rk->rk_metadata_cache.rkmc_cnd_lock); |
| 546 | |
| 547 | #if ENABLE_DEVEL |
| 548 | rd_kafka_dbg(rk, METADATA, "CACHEWAIT" , |
| 549 | "%s wait took %dms: %s" , |
| 550 | __FUNCTION__, (int)((rd_clock() - ts_start)/1000), |
| 551 | r == thrd_success ? "succeeded" : "timed out" ); |
| 552 | #endif |
| 553 | return r == thrd_success; |
| 554 | } |
| 555 | |
| 556 | /** |
| 557 | * @brief Propagate that the cache changed (but not what changed) to |
| 558 | * any cnd listeners. |
| 559 | * @locks none |
| 560 | * @locality any |
| 561 | */ |
| 562 | static void rd_kafka_metadata_cache_propagate_changes (rd_kafka_t *rk) { |
| 563 | mtx_lock(&rk->rk_metadata_cache.rkmc_cnd_lock); |
| 564 | cnd_broadcast(&rk->rk_metadata_cache.rkmc_cnd); |
| 565 | mtx_unlock(&rk->rk_metadata_cache.rkmc_cnd_lock); |
| 566 | } |
| 567 | |
| 568 | /** |
| 569 | * @returns the shared metadata for a topic, or NULL if not found in |
| 570 | * cache. |
| 571 | * |
| 572 | * @locks rd_kafka_*lock() |
| 573 | */ |
| 574 | const rd_kafka_metadata_topic_t * |
| 575 | rd_kafka_metadata_cache_topic_get (rd_kafka_t *rk, const char *topic, |
| 576 | int valid) { |
| 577 | struct rd_kafka_metadata_cache_entry *rkmce; |
| 578 | |
| 579 | if (!(rkmce = rd_kafka_metadata_cache_find(rk, topic, valid))) |
| 580 | return NULL; |
| 581 | |
| 582 | return &rkmce->rkmce_mtopic; |
| 583 | } |
| 584 | |
| 585 | |
| 586 | |
| 587 | |
| 588 | /** |
| 589 | * @brief Looks up the shared metadata for a partition along with its topic. |
| 590 | * |
| 591 | * @param mtopicp: pointer to topic metadata |
| 592 | * @param mpartp: pointer to partition metadata |
| 593 | * @param valid: only return valid entries (no hints) |
| 594 | * |
| 595 | * @returns -1 if topic was not found in cache, 0 if topic was found |
| 596 | * but not the partition, 1 if both topic and partition was found. |
| 597 | * |
| 598 | * @locks rd_kafka_*lock() |
| 599 | */ |
| 600 | int rd_kafka_metadata_cache_topic_partition_get ( |
| 601 | rd_kafka_t *rk, |
| 602 | const rd_kafka_metadata_topic_t **mtopicp, |
| 603 | const rd_kafka_metadata_partition_t **mpartp, |
| 604 | const char *topic, int32_t partition, int valid) { |
| 605 | |
| 606 | const rd_kafka_metadata_topic_t *mtopic; |
| 607 | const rd_kafka_metadata_partition_t *mpart; |
| 608 | rd_kafka_metadata_partition_t skel = { .id = partition }; |
| 609 | |
| 610 | *mtopicp = NULL; |
| 611 | *mpartp = NULL; |
| 612 | |
| 613 | if (!(mtopic = rd_kafka_metadata_cache_topic_get(rk, topic, valid))) |
| 614 | return -1; |
| 615 | |
| 616 | *mtopicp = mtopic; |
| 617 | |
| 618 | /* Partitions array may be sparse so use bsearch lookup. */ |
| 619 | mpart = bsearch(&skel, mtopic->partitions, |
| 620 | mtopic->partition_cnt, |
| 621 | sizeof(*mtopic->partitions), |
| 622 | rd_kafka_metadata_partition_id_cmp); |
| 623 | |
| 624 | if (!mpart) |
| 625 | return 0; |
| 626 | |
| 627 | *mpartp = mpart; |
| 628 | |
| 629 | return 1; |
| 630 | } |
| 631 | |
| 632 | |
| 633 | /** |
| 634 | * @returns the number of topics in \p topics that are in the cache. |
| 635 | * |
| 636 | * @param topics rd_list(const char *): topic names |
| 637 | * @param metadata_agep: age of oldest entry will be returned. |
| 638 | * |
| 639 | * @locks rd_kafka_*lock() |
| 640 | */ |
| 641 | int rd_kafka_metadata_cache_topics_count_exists (rd_kafka_t *rk, |
| 642 | const rd_list_t *topics, |
| 643 | int *metadata_agep) { |
| 644 | const char *topic; |
| 645 | int i; |
| 646 | int cnt = 0; |
| 647 | int max_age = -1; |
| 648 | |
| 649 | RD_LIST_FOREACH(topic, topics, i) { |
| 650 | const struct rd_kafka_metadata_cache_entry *rkmce; |
| 651 | int age; |
| 652 | |
| 653 | if (!(rkmce = rd_kafka_metadata_cache_find(rk, topic, |
| 654 | 1/*valid only*/))) |
| 655 | continue; |
| 656 | |
| 657 | age = (int)((rd_clock() - rkmce->rkmce_ts_insert)/1000); |
| 658 | if (age > max_age) |
| 659 | max_age = age; |
| 660 | cnt++; |
| 661 | } |
| 662 | |
| 663 | *metadata_agep = max_age; |
| 664 | |
| 665 | return cnt; |
| 666 | |
| 667 | } |
| 668 | |
| 669 | |
| 670 | /** |
| 671 | * @brief Copies any topics in \p src to \p dst that have a valid cache |
| 672 | * entry, or not in the cache at all. |
| 673 | * |
| 674 | * In other words; hinted non-valid topics will not copied to \p dst. |
| 675 | * |
| 676 | * @returns the number of topics copied |
| 677 | * |
| 678 | * @locks rd_kafka_*lock() |
| 679 | */ |
| 680 | int rd_kafka_metadata_cache_topics_filter_hinted (rd_kafka_t *rk, |
| 681 | rd_list_t *dst, |
| 682 | const rd_list_t *src) { |
| 683 | const char *topic; |
| 684 | int i; |
| 685 | int cnt = 0; |
| 686 | |
| 687 | |
| 688 | RD_LIST_FOREACH(topic, src, i) { |
| 689 | const struct rd_kafka_metadata_cache_entry *rkmce; |
| 690 | |
| 691 | rkmce = rd_kafka_metadata_cache_find(rk, topic, 0/*any sort*/); |
| 692 | if (rkmce && !RD_KAFKA_METADATA_CACHE_VALID(rkmce)) |
| 693 | continue; |
| 694 | |
| 695 | rd_list_add(dst, rd_strdup(topic)); |
| 696 | cnt++; |
| 697 | } |
| 698 | |
| 699 | return cnt; |
| 700 | } |
| 701 | |
| 702 | |
| 703 | |
| 704 | /** |
| 705 | * @brief Dump cache to \p fp |
| 706 | * |
| 707 | * @locks rd_kafka_*lock() |
| 708 | */ |
| 709 | void rd_kafka_metadata_cache_dump (FILE *fp, rd_kafka_t *rk) { |
| 710 | const struct rd_kafka_metadata_cache *rkmc = &rk->rk_metadata_cache; |
| 711 | const struct rd_kafka_metadata_cache_entry *rkmce; |
| 712 | rd_ts_t now = rd_clock(); |
| 713 | |
| 714 | fprintf(fp, |
| 715 | "Metadata cache with %d entries:\n" , |
| 716 | rkmc->rkmc_cnt); |
| 717 | TAILQ_FOREACH(rkmce, &rkmc->rkmc_expiry, rkmce_link) { |
| 718 | fprintf(fp, |
| 719 | " %s (inserted %dms ago, expires in %dms, " |
| 720 | "%d partition(s), %s)%s%s\n" , |
| 721 | rkmce->rkmce_mtopic.topic, |
| 722 | (int)((now - rkmce->rkmce_ts_insert)/1000), |
| 723 | (int)((rkmce->rkmce_ts_expires - now)/1000), |
| 724 | rkmce->rkmce_mtopic.partition_cnt, |
| 725 | RD_KAFKA_METADATA_CACHE_VALID(rkmce) ? "valid" :"hint" , |
| 726 | rkmce->rkmce_mtopic.err ? " error: " : "" , |
| 727 | rkmce->rkmce_mtopic.err ? |
| 728 | rd_kafka_err2str(rkmce->rkmce_mtopic.err) : "" ); |
| 729 | } |
| 730 | } |
| 731 | |
| 732 | /**@}*/ |
| 733 | |