2 * Copyright (C) 1996-2022 The Squid Software Foundation and contributors
4 * Squid software is distributed under GPLv2+ license and includes
5 * contributions from numerous individuals and organizations.
6 * Please see the COPYING and CONTRIBUTORS files for details.
9 /* DEBUG: section 71 Store Digest Manager */
12 * TODO: We probably do not track all the cases when
13 * storeDigestNoteStoreReady() must be called; this may prevent
14 * storeDigestRebuild/write schedule to be activated
18 #include "debug/Stream.h"
21 #include "mgr/Registration.h"
22 #include "store_digest.h"
25 #include "CacheDigest.h"
26 #include "HttpReply.h"
27 #include "HttpRequest.h"
29 #include "MemObject.h"
30 #include "PeerDigest.h"
32 #include "SquidConfig.h"
34 #include "StoreSearch.h"
43 class StoreDigestState
46 StoreDigestCBlock cblock
;
47 int rebuild_lock
= 0; ///< bucket number
48 StoreEntry
* rewrite_lock
= nullptr; ///< points to store entry with the digest
49 StoreEntry
* publicEntry
= nullptr; ///< points to the previous store entry with the digest
50 StoreSearchPointer theSearch
;
51 int rewrite_offset
= 0;
52 int rebuild_count
= 0;
53 int rewrite_count
= 0;
56 class StoreDigestStats
59 int del_count
= 0; /* #store entries deleted from store_digest */
60 int del_lost_count
= 0; /* #store entries not found in store_digest on delete */
61 int add_count
= 0; /* #store entries accepted to store_digest */
62 int add_coll_count
= 0; /* #accepted entries that collided with existing ones */
63 int rej_count
= 0; /* #store entries not accepted to store_digest */
64 int rej_coll_count
= 0; /* #not accepted entries that collided with existing ones */
68 static StoreDigestState sd_state
;
69 static StoreDigestStats sd_stats
;
71 /* local prototypes */
72 static void storeDigestRebuildStart(void *datanotused
);
73 static void storeDigestRebuildResume(void);
74 static void storeDigestRebuildFinish(void);
75 static void storeDigestRebuildStep(void *datanotused
);
76 static void storeDigestRewriteStart(void *);
77 static void storeDigestRewriteResume(void);
78 static void storeDigestRewriteFinish(StoreEntry
* e
);
79 static EVH storeDigestSwapOutStep
;
80 static void storeDigestCBlockSwapOut(StoreEntry
* e
);
81 static void storeDigestAdd(const StoreEntry
*);
83 /// calculates digest capacity
88 * To-Do: Bloom proved that the optimal filter utilization is 50% (half of
89 * the bits are off). However, we do not have a formula to calculate the
90 * number of _entries_ we want to pre-allocate for.
92 const uint64_t hi_cap
= Store::Root().maxSize() / Config
.Store
.avgObjectSize
;
93 const uint64_t lo_cap
= 1 + Store::Root().currentSize() / Config
.Store
.avgObjectSize
;
94 const uint64_t e_count
= StoreEntry::inUseCount();
95 uint64_t cap
= e_count
? e_count
: hi_cap
;
96 debugs(71, 2, "have: " << e_count
<< ", want " << cap
<<
97 " entries; limits: [" << lo_cap
<< ", " << hi_cap
<< "]");
102 /* do not enforce hi_cap limit, average-based estimation may be wrong
107 // Bug 4534: we still have to set an upper-limit at some reasonable value though.
108 // this matches cacheDigestCalcMaskSize doing (cap*bpe)+7 < INT_MAX
109 const uint64_t absolute_max
= (INT_MAX
-8) / Config
.digest
.bits_per_entry
;
110 if (cap
> absolute_max
) {
111 static time_t last_loud
= 0;
112 if (last_loud
< squid_curtime
- 86400) {
113 debugs(71, DBG_IMPORTANT
, "WARNING: Cache Digest cannot store " << cap
<< " entries. Limiting to " << absolute_max
);
114 last_loud
= squid_curtime
;
116 debugs(71, 3, "WARNING: Cache Digest cannot store " << cap
<< " entries. Limiting to " << absolute_max
);
123 #endif /* USE_CACHE_DIGESTS */
126 storeDigestInit(void)
128 Mgr::RegisterAction("store_digest", "Store Digest", storeDigestReport
, 0, 1);
130 #if USE_CACHE_DIGESTS
131 if (!Config
.onoff
.digest_generation
) {
133 debugs(71, 3, "Local cache digest generation disabled");
137 const uint64_t cap
= storeDigestCalcCap();
138 store_digest
= new CacheDigest(cap
, Config
.digest
.bits_per_entry
);
139 debugs(71, DBG_IMPORTANT
, "Local cache digest enabled; rebuild/rewrite every " <<
140 (int) Config
.digest
.rebuild_period
<< "/" <<
141 (int) Config
.digest
.rewrite_period
<< " sec");
143 sd_state
= StoreDigestState();
146 debugs(71, 3, "Local cache digest is 'off'");
150 /* called when store_rebuild completes */
152 storeDigestNoteStoreReady(void)
154 #if USE_CACHE_DIGESTS
156 if (Config
.onoff
.digest_generation
) {
157 storeDigestRebuildStart(NULL
);
158 storeDigestRewriteStart(NULL
);
164 //TODO: this seems to be dead code. Is it needed?
166 storeDigestDel(const StoreEntry
* entry
)
168 #if USE_CACHE_DIGESTS
170 if (!Config
.onoff
.digest_generation
) {
174 assert(entry
&& store_digest
);
175 debugs(71, 6, "storeDigestDel: checking entry, key: " << entry
->getMD5Text());
177 if (!EBIT_TEST(entry
->flags
, KEY_PRIVATE
)) {
178 if (!store_digest
->contains(static_cast<const cache_key
*>(entry
->key
))) {
179 ++sd_stats
.del_lost_count
;
180 debugs(71, 6, "storeDigestDel: lost entry, key: " << entry
->getMD5Text() << " url: " << entry
->url() );
182 ++sd_stats
.del_count
;
183 store_digest
->remove(static_cast<const cache_key
*>(entry
->key
));
184 debugs(71, 6, "storeDigestDel: deled entry, key: " << entry
->getMD5Text());
189 #endif //USE_CACHE_DIGESTS
193 storeDigestReport(StoreEntry
* e
)
195 #if USE_CACHE_DIGESTS
197 if (!Config
.onoff
.digest_generation
) {
202 static const SBuf
label("store");
203 cacheDigestReport(store_digest
, label
, e
);
204 storeAppendPrintf(e
, "\t added: %d rejected: %d ( %.2f %%) del-ed: %d\n",
207 xpercent(sd_stats
.rej_count
, sd_stats
.rej_count
+ sd_stats
.add_count
),
209 storeAppendPrintf(e
, "\t collisions: on add: %.2f %% on rej: %.2f %%\n",
210 xpercent(sd_stats
.add_coll_count
, sd_stats
.add_count
),
211 xpercent(sd_stats
.rej_coll_count
, sd_stats
.rej_count
));
213 storeAppendPrintf(e
, "store digest: disabled.\n");
217 #endif //USE_CACHE_DIGESTS
224 #if USE_CACHE_DIGESTS
226 /* should we digest this entry? used by storeDigestAdd() */
228 storeDigestAddable(const StoreEntry
* e
)
230 /* add some stats! XXX */
232 debugs(71, 6, "storeDigestAddable: checking entry, key: " << e
->getMD5Text());
234 /* check various entry flags (mimics StoreEntry::checkCachable XXX) */
236 if (EBIT_TEST(e
->flags
, KEY_PRIVATE
)) {
237 debugs(71, 6, "storeDigestAddable: NO: private key");
241 if (EBIT_TEST(e
->flags
, ENTRY_NEGCACHED
)) {
242 debugs(71, 6, "storeDigestAddable: NO: negative cached");
246 if (EBIT_TEST(e
->flags
, RELEASE_REQUEST
)) {
247 debugs(71, 6, "storeDigestAddable: NO: release requested");
251 if (e
->store_status
== STORE_OK
&& EBIT_TEST(e
->flags
, ENTRY_BAD_LENGTH
)) {
252 debugs(71, 6, "storeDigestAddable: NO: wrong content-length");
256 /* do not digest huge objects */
257 if (e
->swap_file_sz
> (uint64_t )Config
.Store
.maxObjectSize
) {
258 debugs(71, 6, "storeDigestAddable: NO: too big");
262 /* still here? check staleness */
263 /* Note: We should use the time of the next rebuild, not (cur_time+period) */
264 if (refreshCheckDigest(e
, Config
.digest
.rebuild_period
)) {
265 debugs(71, 6, "storeDigestAdd: entry expires within " << Config
.digest
.rebuild_period
<< " secs, ignoring");
270 * idea: how about also skipping very fresh (thus, potentially
271 * unstable) entries? Should be configurable through
272 * cd_refresh_pattern, of course.
275 * idea: skip objects that are going to be purged before the next
282 storeDigestAdd(const StoreEntry
* entry
)
284 assert(entry
&& store_digest
);
286 if (storeDigestAddable(entry
)) {
287 ++sd_stats
.add_count
;
289 if (store_digest
->contains(static_cast<const cache_key
*>(entry
->key
)))
290 ++sd_stats
.add_coll_count
;
292 store_digest
->add(static_cast<const cache_key
*>(entry
->key
));
294 debugs(71, 6, "storeDigestAdd: added entry, key: " << entry
->getMD5Text());
296 ++sd_stats
.rej_count
;
298 if (store_digest
->contains(static_cast<const cache_key
*>(entry
->key
)))
299 ++sd_stats
.rej_coll_count
;
303 /* rebuilds digest from scratch */
305 storeDigestRebuildStart(void *)
307 assert(store_digest
);
308 /* prevent overlapping if rebuild schedule is too tight */
310 if (sd_state
.rebuild_lock
) {
311 debugs(71, DBG_IMPORTANT
, "storeDigestRebuildStart: overlap detected, consider increasing rebuild period");
315 sd_state
.rebuild_lock
= 1;
316 debugs(71, 2, "storeDigestRebuildStart: rebuild #" << sd_state
.rebuild_count
+ 1);
318 if (sd_state
.rewrite_lock
) {
319 debugs(71, 2, "storeDigestRebuildStart: waiting for Rewrite to finish.");
323 storeDigestRebuildResume();
326 /// \returns true if we actually resized the digest
330 const uint64_t cap
= storeDigestCalcCap();
331 assert(store_digest
);
333 if (cap
> store_digest
->capacity
)
334 diff
= cap
- store_digest
->capacity
;
336 diff
= store_digest
->capacity
- cap
;
337 debugs(71, 2, store_digest
->capacity
<< " -> " << cap
<< "; change: " <<
338 diff
<< " (" << xpercentInt(diff
, store_digest
->capacity
) << "%)" );
339 /* avoid minor adjustments */
341 if (diff
<= store_digest
->capacity
/ 10) {
342 debugs(71, 2, "small change, will not resize.");
345 debugs(71, 2, "big change, resizing.");
346 store_digest
->updateCapacity(cap
);
351 /* called be Rewrite to push Rebuild forward */
353 storeDigestRebuildResume(void)
355 assert(sd_state
.rebuild_lock
);
356 assert(!sd_state
.rewrite_lock
);
357 sd_state
.theSearch
= Store::Root().search();
358 /* resize or clear */
360 if (!storeDigestResize())
361 store_digest
->clear(); /* not clean()! */
363 sd_stats
= StoreDigestStats();
365 eventAdd("storeDigestRebuildStep", storeDigestRebuildStep
, NULL
, 0.0, 1);
368 /* finishes swap out sequence for the digest; schedules next rebuild */
370 storeDigestRebuildFinish(void)
372 assert(sd_state
.rebuild_lock
);
373 sd_state
.rebuild_lock
= 0;
374 ++sd_state
.rebuild_count
;
375 debugs(71, 2, "storeDigestRebuildFinish: done.");
376 eventAdd("storeDigestRebuildStart", storeDigestRebuildStart
, NULL
, (double)
377 Config
.digest
.rebuild_period
, 1);
378 /* resume pending Rewrite if any */
380 if (sd_state
.rewrite_lock
)
381 storeDigestRewriteResume();
384 /* recalculate a few hash buckets per invocation; schedules next step */
386 storeDigestRebuildStep(void *)
388 /* TODO: call Store::Root().size() to determine this.. */
389 int count
= Config
.Store
.objectsPerBucket
* (int) ceil((double) store_hash_buckets
*
390 (double) Config
.digest
.rebuild_chunk_percentage
/ 100.0);
391 assert(sd_state
.rebuild_lock
);
393 debugs(71, 3, "storeDigestRebuildStep: buckets: " << store_hash_buckets
<< " entries to check: " << count
);
395 while (count
-- && !sd_state
.theSearch
->isDone() && sd_state
.theSearch
->next())
396 storeDigestAdd(sd_state
.theSearch
->currentItem());
399 if (sd_state
.theSearch
->isDone())
400 storeDigestRebuildFinish();
402 eventAdd("storeDigestRebuildStep", storeDigestRebuildStep
, NULL
, 0.0, 1);
405 /* starts swap out sequence for the digest */
407 storeDigestRewriteStart(void *)
409 assert(store_digest
);
410 /* prevent overlapping if rewrite schedule is too tight */
412 if (sd_state
.rewrite_lock
) {
413 debugs(71, DBG_IMPORTANT
, "storeDigestRewrite: overlap detected, consider increasing rewrite period");
417 debugs(71, 2, "storeDigestRewrite: start rewrite #" << sd_state
.rewrite_count
+ 1);
419 const char *url
= internalLocalUri("/squid-internal-periodic/", SBuf(StoreDigestFileName
));
420 const auto mx
= MasterXaction::MakePortless
<XactionInitiator::initCacheDigest
>();
421 auto req
= HttpRequest::FromUrlXXX(url
, mx
);
424 flags
.cachable
= true;
426 StoreEntry
*e
= storeCreateEntry(url
, url
, flags
, Http::METHOD_GET
);
428 sd_state
.rewrite_lock
= e
;
429 debugs(71, 3, "storeDigestRewrite: url: " << url
<< " key: " << e
->getMD5Text());
430 e
->mem_obj
->request
= req
;
432 /* wait for rebuild (if any) to finish */
433 if (sd_state
.rebuild_lock
) {
434 debugs(71, 2, "storeDigestRewriteStart: waiting for rebuild to finish.");
438 storeDigestRewriteResume();
442 storeDigestRewriteResume(void)
446 assert(sd_state
.rewrite_lock
);
447 assert(!sd_state
.rebuild_lock
);
448 e
= sd_state
.rewrite_lock
;
449 sd_state
.rewrite_offset
= 0;
450 EBIT_SET(e
->flags
, ENTRY_SPECIAL
);
451 /* setting public key will mark the old digest entry for removal once unlocked */
453 if (const auto oldEntry
= sd_state
.publicEntry
) {
454 oldEntry
->release(true);
455 sd_state
.publicEntry
= nullptr;
456 oldEntry
->unlock("storeDigestRewriteResume");
459 sd_state
.publicEntry
= e
;
461 HttpReply
*rep
= new HttpReply
;
462 rep
->setHeaders(Http::scOkay
, "Cache Digest OK",
463 "application/cache-digest", (store_digest
->mask_size
+ sizeof(sd_state
.cblock
)),
464 squid_curtime
, (squid_curtime
+ Config
.digest
.rewrite_period
) );
465 debugs(71, 3, "storeDigestRewrite: entry expires on " << rep
->expires
<<
466 " (" << std::showpos
<< (int) (rep
->expires
- squid_curtime
) << ")");
468 e
->replaceHttpReply(rep
);
469 storeDigestCBlockSwapOut(e
);
471 eventAdd("storeDigestSwapOutStep", storeDigestSwapOutStep
, sd_state
.rewrite_lock
, 0.0, 1, false);
474 /* finishes swap out sequence for the digest; schedules next rewrite */
476 storeDigestRewriteFinish(StoreEntry
* e
)
478 assert(e
== sd_state
.rewrite_lock
);
481 debugs(71, 2, "storeDigestRewriteFinish: digest expires at " << e
->expires
<<
482 " (" << std::showpos
<< (int) (e
->expires
- squid_curtime
) << ")");
483 /* is this the write order? @?@ */
484 e
->mem_obj
->unlinkRequest();
485 sd_state
.rewrite_lock
= NULL
;
486 ++sd_state
.rewrite_count
;
487 eventAdd("storeDigestRewriteStart", storeDigestRewriteStart
, NULL
, (double)
488 Config
.digest
.rewrite_period
, 1);
489 /* resume pending Rebuild if any */
491 if (sd_state
.rebuild_lock
)
492 storeDigestRebuildResume();
495 /* swaps out one digest "chunk" per invocation; schedules next swap out */
497 storeDigestSwapOutStep(void *data
)
499 StoreEntry
*e
= static_cast<StoreEntry
*>(data
);
500 int chunk_size
= Config
.digest
.swapout_chunk_size
;
501 assert(e
== sd_state
.rewrite_lock
);
503 /* _add_ check that nothing bad happened while we were waiting @?@ @?@ */
505 if (static_cast<uint32_t>(sd_state
.rewrite_offset
+ chunk_size
) > store_digest
->mask_size
)
506 chunk_size
= store_digest
->mask_size
- sd_state
.rewrite_offset
;
508 e
->append(store_digest
->mask
+ sd_state
.rewrite_offset
, chunk_size
);
510 debugs(71, 3, "storeDigestSwapOutStep: size: " << store_digest
->mask_size
<<
511 " offset: " << sd_state
.rewrite_offset
<< " chunk: " <<
512 chunk_size
<< " bytes");
514 sd_state
.rewrite_offset
+= chunk_size
;
517 if (static_cast<uint32_t>(sd_state
.rewrite_offset
) >= store_digest
->mask_size
)
518 storeDigestRewriteFinish(e
);
520 eventAdd("storeDigestSwapOutStep", storeDigestSwapOutStep
, data
, 0.0, 1, false);
524 storeDigestCBlockSwapOut(StoreEntry
* e
)
526 memset(&sd_state
.cblock
, 0, sizeof(sd_state
.cblock
));
527 sd_state
.cblock
.ver
.current
= htons(CacheDigestVer
.current
);
528 sd_state
.cblock
.ver
.required
= htons(CacheDigestVer
.required
);
529 sd_state
.cblock
.capacity
= htonl(store_digest
->capacity
);
530 sd_state
.cblock
.count
= htonl(store_digest
->count
);
531 sd_state
.cblock
.del_count
= htonl(store_digest
->del_count
);
532 sd_state
.cblock
.mask_size
= htonl(store_digest
->mask_size
);
533 sd_state
.cblock
.bits_per_entry
= Config
.digest
.bits_per_entry
;
534 sd_state
.cblock
.hash_func_count
= (unsigned char) CacheDigestHashFuncCount
;
535 e
->append((char *) &sd_state
.cblock
, sizeof(sd_state
.cblock
));
538 #endif /* USE_CACHE_DIGESTS */