/* * Copyright (C) 1996-2023 The Squid Software Foundation and contributors * * Squid software is distributed under GPLv2+ license and includes * contributions from numerous individuals and organizations. * Please see the COPYING and CONTRIBUTORS files for details. */ /* DEBUG: section 71 Store Digest Manager */ /* * TODO: We probably do not track all the cases when * storeDigestNoteStoreReady() must be called; this may prevent * storeDigestRebuild/write schedule to be activated */ #include "squid.h" #include "debug/Stream.h" #include "event.h" #include "globals.h" #include "mgr/Registration.h" #include "store_digest.h" #if USE_CACHE_DIGESTS #include "CacheDigest.h" #include "HttpReply.h" #include "HttpRequest.h" #include "internal.h" #include "MemObject.h" #include "PeerDigest.h" #include "refresh.h" #include "SquidConfig.h" #include "Store.h" #include "StoreSearch.h" #include "util.h" #include /* * local types */ class StoreDigestState { public: StoreDigestCBlock cblock; int rebuild_lock = 0; ///< bucket number StoreEntry * rewrite_lock = nullptr; ///< points to store entry with the digest StoreEntry * publicEntry = nullptr; ///< points to the previous store entry with the digest StoreSearchPointer theSearch; int rewrite_offset = 0; int rebuild_count = 0; int rewrite_count = 0; }; class StoreDigestStats { public: int del_count = 0; /* #store entries deleted from store_digest */ int del_lost_count = 0; /* #store entries not found in store_digest on delete */ int add_count = 0; /* #store entries accepted to store_digest */ int add_coll_count = 0; /* #accepted entries that collided with existing ones */ int rej_count = 0; /* #store entries not accepted to store_digest */ int rej_coll_count = 0; /* #not accepted entries that collided with existing ones */ }; /* local vars */ static StoreDigestState sd_state; static StoreDigestStats sd_stats; /* local prototypes */ static void storeDigestRebuildStart(void *datanotused); static void storeDigestRebuildResume(void); static void storeDigestRebuildFinish(void); static void storeDigestRebuildStep(void *datanotused); static void storeDigestRewriteStart(void *); static void storeDigestRewriteResume(void); static void storeDigestRewriteFinish(StoreEntry * e); static EVH storeDigestSwapOutStep; static void storeDigestCBlockSwapOut(StoreEntry * e); static void storeDigestAdd(const StoreEntry *); /// calculates digest capacity static uint64_t storeDigestCalcCap() { /* * To-Do: Bloom proved that the optimal filter utilization is 50% (half of * the bits are off). However, we do not have a formula to calculate the * number of _entries_ we want to pre-allocate for. */ const uint64_t hi_cap = Store::Root().maxSize() / Config.Store.avgObjectSize; const uint64_t lo_cap = 1 + Store::Root().currentSize() / Config.Store.avgObjectSize; const uint64_t e_count = StoreEntry::inUseCount(); uint64_t cap = e_count ? e_count : hi_cap; debugs(71, 2, "have: " << e_count << ", want " << cap << " entries; limits: [" << lo_cap << ", " << hi_cap << "]"); if (cap < lo_cap) cap = lo_cap; /* do not enforce hi_cap limit, average-based estimation may be wrong *if (cap > hi_cap) * cap = hi_cap; */ // Bug 4534: we still have to set an upper-limit at some reasonable value though. // this matches cacheDigestCalcMaskSize doing (cap*bpe)+7 < INT_MAX const uint64_t absolute_max = (INT_MAX -8) / Config.digest.bits_per_entry; if (cap > absolute_max) { static time_t last_loud = 0; if (last_loud < squid_curtime - 86400) { debugs(71, DBG_IMPORTANT, "WARNING: Cache Digest cannot store " << cap << " entries. Limiting to " << absolute_max); last_loud = squid_curtime; } else { debugs(71, 3, "WARNING: Cache Digest cannot store " << cap << " entries. Limiting to " << absolute_max); } cap = absolute_max; } return cap; } #endif /* USE_CACHE_DIGESTS */ void storeDigestInit(void) { Mgr::RegisterAction("store_digest", "Store Digest", storeDigestReport, 0, 1); #if USE_CACHE_DIGESTS if (!Config.onoff.digest_generation) { store_digest = nullptr; debugs(71, 3, "Local cache digest generation disabled"); return; } const uint64_t cap = storeDigestCalcCap(); store_digest = new CacheDigest(cap, Config.digest.bits_per_entry); debugs(71, DBG_IMPORTANT, "Local cache digest enabled; rebuild/rewrite every " << (int) Config.digest.rebuild_period << "/" << (int) Config.digest.rewrite_period << " sec"); sd_state = StoreDigestState(); #else store_digest = nullptr; debugs(71, 3, "Local cache digest is 'off'"); #endif } /* called when store_rebuild completes */ void storeDigestNoteStoreReady(void) { #if USE_CACHE_DIGESTS if (Config.onoff.digest_generation) { storeDigestRebuildStart(nullptr); storeDigestRewriteStart(nullptr); } #endif } //TODO: this seems to be dead code. Is it needed? void storeDigestDel(const StoreEntry * entry) { #if USE_CACHE_DIGESTS if (!Config.onoff.digest_generation) { return; } assert(entry && store_digest); debugs(71, 6, "storeDigestDel: checking entry, key: " << entry->getMD5Text()); if (!EBIT_TEST(entry->flags, KEY_PRIVATE)) { if (!store_digest->contains(static_cast(entry->key))) { ++sd_stats.del_lost_count; debugs(71, 6, "storeDigestDel: lost entry, key: " << entry->getMD5Text() << " url: " << entry->url() ); } else { ++sd_stats.del_count; store_digest->remove(static_cast(entry->key)); debugs(71, 6, "storeDigestDel: deled entry, key: " << entry->getMD5Text()); } } #else (void)entry; #endif //USE_CACHE_DIGESTS } void storeDigestReport(StoreEntry * e) { #if USE_CACHE_DIGESTS if (!Config.onoff.digest_generation) { return; } if (store_digest) { static const SBuf label("store"); cacheDigestReport(store_digest, label, e); storeAppendPrintf(e, "\t added: %d rejected: %d ( %.2f %%) del-ed: %d\n", sd_stats.add_count, sd_stats.rej_count, xpercent(sd_stats.rej_count, sd_stats.rej_count + sd_stats.add_count), sd_stats.del_count); storeAppendPrintf(e, "\t collisions: on add: %.2f %% on rej: %.2f %%\n", xpercent(sd_stats.add_coll_count, sd_stats.add_count), xpercent(sd_stats.rej_coll_count, sd_stats.rej_count)); } else { storeAppendPrintf(e, "store digest: disabled.\n"); } #else (void)e; #endif //USE_CACHE_DIGESTS } /* * LOCAL FUNCTIONS */ #if USE_CACHE_DIGESTS /* should we digest this entry? used by storeDigestAdd() */ static int storeDigestAddable(const StoreEntry * e) { /* add some stats! XXX */ debugs(71, 6, "storeDigestAddable: checking entry, key: " << e->getMD5Text()); /* check various entry flags (mimics StoreEntry::checkCachable XXX) */ if (EBIT_TEST(e->flags, KEY_PRIVATE)) { debugs(71, 6, "storeDigestAddable: NO: private key"); return 0; } if (EBIT_TEST(e->flags, ENTRY_NEGCACHED)) { debugs(71, 6, "storeDigestAddable: NO: negative cached"); return 0; } if (EBIT_TEST(e->flags, RELEASE_REQUEST)) { debugs(71, 6, "storeDigestAddable: NO: release requested"); return 0; } if (EBIT_TEST(e->flags, ENTRY_BAD_LENGTH)) { debugs(71, 6, "storeDigestAddable: NO: wrong content-length"); return 0; } /* do not digest huge objects */ if (e->swap_file_sz > (uint64_t )Config.Store.maxObjectSize) { debugs(71, 6, "storeDigestAddable: NO: too big"); return 0; } /* still here? check staleness */ /* Note: We should use the time of the next rebuild, not (cur_time+period) */ if (refreshCheckDigest(e, Config.digest.rebuild_period)) { debugs(71, 6, "storeDigestAdd: entry expires within " << Config.digest.rebuild_period << " secs, ignoring"); return 0; } /* * idea: how about also skipping very fresh (thus, potentially * unstable) entries? Should be configurable through * cd_refresh_pattern, of course. */ /* * idea: skip objects that are going to be purged before the next * update. */ return 1; } static void storeDigestAdd(const StoreEntry * entry) { assert(entry && store_digest); if (storeDigestAddable(entry)) { ++sd_stats.add_count; if (store_digest->contains(static_cast(entry->key))) ++sd_stats.add_coll_count; store_digest->add(static_cast(entry->key)); debugs(71, 6, "storeDigestAdd: added entry, key: " << entry->getMD5Text()); } else { ++sd_stats.rej_count; if (store_digest->contains(static_cast(entry->key))) ++sd_stats.rej_coll_count; } } /* rebuilds digest from scratch */ static void storeDigestRebuildStart(void *) { assert(store_digest); /* prevent overlapping if rebuild schedule is too tight */ if (sd_state.rebuild_lock) { debugs(71, DBG_IMPORTANT, "storeDigestRebuildStart: overlap detected, consider increasing rebuild period"); return; } sd_state.rebuild_lock = 1; debugs(71, 2, "storeDigestRebuildStart: rebuild #" << sd_state.rebuild_count + 1); if (sd_state.rewrite_lock) { debugs(71, 2, "storeDigestRebuildStart: waiting for Rewrite to finish."); return; } storeDigestRebuildResume(); } /// \returns true if we actually resized the digest static bool storeDigestResize() { const uint64_t cap = storeDigestCalcCap(); assert(store_digest); uint64_t diff; if (cap > store_digest->capacity) diff = cap - store_digest->capacity; else diff = store_digest->capacity - cap; debugs(71, 2, store_digest->capacity << " -> " << cap << "; change: " << diff << " (" << xpercentInt(diff, store_digest->capacity) << "%)" ); /* avoid minor adjustments */ if (diff <= store_digest->capacity / 10) { debugs(71, 2, "small change, will not resize."); return false; } else { debugs(71, 2, "big change, resizing."); store_digest->updateCapacity(cap); } return true; } /* called be Rewrite to push Rebuild forward */ static void storeDigestRebuildResume(void) { assert(sd_state.rebuild_lock); assert(!sd_state.rewrite_lock); sd_state.theSearch = Store::Root().search(); /* resize or clear */ if (!storeDigestResize()) store_digest->clear(); /* not clean()! */ sd_stats = StoreDigestStats(); eventAdd("storeDigestRebuildStep", storeDigestRebuildStep, nullptr, 0.0, 1); } /* finishes swap out sequence for the digest; schedules next rebuild */ static void storeDigestRebuildFinish(void) { assert(sd_state.rebuild_lock); sd_state.rebuild_lock = 0; ++sd_state.rebuild_count; debugs(71, 2, "storeDigestRebuildFinish: done."); eventAdd("storeDigestRebuildStart", storeDigestRebuildStart, nullptr, (double) Config.digest.rebuild_period, 1); /* resume pending Rewrite if any */ if (sd_state.rewrite_lock) storeDigestRewriteResume(); } /* recalculate a few hash buckets per invocation; schedules next step */ static void storeDigestRebuildStep(void *) { /* TODO: call Store::Root().size() to determine this.. */ int count = Config.Store.objectsPerBucket * (int) ceil((double) store_hash_buckets * (double) Config.digest.rebuild_chunk_percentage / 100.0); assert(sd_state.rebuild_lock); debugs(71, 3, "storeDigestRebuildStep: buckets: " << store_hash_buckets << " entries to check: " << count); while (count-- && !sd_state.theSearch->isDone() && sd_state.theSearch->next()) storeDigestAdd(sd_state.theSearch->currentItem()); /* are we done ? */ if (sd_state.theSearch->isDone()) storeDigestRebuildFinish(); else eventAdd("storeDigestRebuildStep", storeDigestRebuildStep, nullptr, 0.0, 1); } /* starts swap out sequence for the digest */ static void storeDigestRewriteStart(void *) { assert(store_digest); /* prevent overlapping if rewrite schedule is too tight */ if (sd_state.rewrite_lock) { debugs(71, DBG_IMPORTANT, "storeDigestRewrite: overlap detected, consider increasing rewrite period"); return; } debugs(71, 2, "storeDigestRewrite: start rewrite #" << sd_state.rewrite_count + 1); const char *url = internalLocalUri("/squid-internal-periodic/", SBuf(StoreDigestFileName)); const auto mx = MasterXaction::MakePortless(); auto req = HttpRequest::FromUrlXXX(url, mx); RequestFlags flags; flags.cachable.support(); // prevent RELEASE_REQUEST in storeCreateEntry() StoreEntry *e = storeCreateEntry(url, url, flags, Http::METHOD_GET); assert(e); sd_state.rewrite_lock = e; debugs(71, 3, "storeDigestRewrite: url: " << url << " key: " << e->getMD5Text()); e->mem_obj->request = req; /* wait for rebuild (if any) to finish */ if (sd_state.rebuild_lock) { debugs(71, 2, "storeDigestRewriteStart: waiting for rebuild to finish."); return; } storeDigestRewriteResume(); } static void storeDigestRewriteResume(void) { StoreEntry *e; assert(sd_state.rewrite_lock); assert(!sd_state.rebuild_lock); e = sd_state.rewrite_lock; sd_state.rewrite_offset = 0; EBIT_SET(e->flags, ENTRY_SPECIAL); /* setting public key will mark the old digest entry for removal once unlocked */ e->setPublicKey(); if (const auto oldEntry = sd_state.publicEntry) { oldEntry->release(true); sd_state.publicEntry = nullptr; oldEntry->unlock("storeDigestRewriteResume"); } assert(e->locked()); sd_state.publicEntry = e; /* fake reply */ HttpReply *rep = new HttpReply; rep->setHeaders(Http::scOkay, "Cache Digest OK", "application/cache-digest", (store_digest->mask_size + sizeof(sd_state.cblock)), squid_curtime, (squid_curtime + Config.digest.rewrite_period) ); debugs(71, 3, "storeDigestRewrite: entry expires on " << rep->expires << " (" << std::showpos << (int) (rep->expires - squid_curtime) << ")"); e->buffer(); e->replaceHttpReply(rep); storeDigestCBlockSwapOut(e); e->flush(); eventAdd("storeDigestSwapOutStep", storeDigestSwapOutStep, sd_state.rewrite_lock, 0.0, 1, false); } /* finishes swap out sequence for the digest; schedules next rewrite */ static void storeDigestRewriteFinish(StoreEntry * e) { assert(e == sd_state.rewrite_lock); e->complete(); e->timestampsSet(); debugs(71, 2, "storeDigestRewriteFinish: digest expires at " << e->expires << " (" << std::showpos << (int) (e->expires - squid_curtime) << ")"); /* is this the write order? @?@ */ e->mem_obj->unlinkRequest(); sd_state.rewrite_lock = nullptr; ++sd_state.rewrite_count; eventAdd("storeDigestRewriteStart", storeDigestRewriteStart, nullptr, (double) Config.digest.rewrite_period, 1); /* resume pending Rebuild if any */ if (sd_state.rebuild_lock) storeDigestRebuildResume(); } /* swaps out one digest "chunk" per invocation; schedules next swap out */ static void storeDigestSwapOutStep(void *data) { StoreEntry *e = static_cast(data); int chunk_size = Config.digest.swapout_chunk_size; assert(e == sd_state.rewrite_lock); assert(e); /* _add_ check that nothing bad happened while we were waiting @?@ @?@ */ if (static_cast(sd_state.rewrite_offset + chunk_size) > store_digest->mask_size) chunk_size = store_digest->mask_size - sd_state.rewrite_offset; e->append(store_digest->mask + sd_state.rewrite_offset, chunk_size); debugs(71, 3, "storeDigestSwapOutStep: size: " << store_digest->mask_size << " offset: " << sd_state.rewrite_offset << " chunk: " << chunk_size << " bytes"); sd_state.rewrite_offset += chunk_size; /* are we done ? */ if (static_cast(sd_state.rewrite_offset) >= store_digest->mask_size) storeDigestRewriteFinish(e); else eventAdd("storeDigestSwapOutStep", storeDigestSwapOutStep, data, 0.0, 1, false); } static void storeDigestCBlockSwapOut(StoreEntry * e) { memset(&sd_state.cblock, 0, sizeof(sd_state.cblock)); sd_state.cblock.ver.current = htons(CacheDigestVer.current); sd_state.cblock.ver.required = htons(CacheDigestVer.required); sd_state.cblock.capacity = htonl(store_digest->capacity); sd_state.cblock.count = htonl(store_digest->count); sd_state.cblock.del_count = htonl(store_digest->del_count); sd_state.cblock.mask_size = htonl(store_digest->mask_size); sd_state.cblock.bits_per_entry = Config.digest.bits_per_entry; sd_state.cblock.hash_func_count = (unsigned char) CacheDigestHashFuncCount; e->append((char *) &sd_state.cblock, sizeof(sd_state.cblock)); } #endif /* USE_CACHE_DIGESTS */