5 * DEBUG: section 20 Store Rebuild Routines
6 * AUTHOR: Duane Wessels
8 * SQUID Web Proxy Cache http://www.squid-cache.org/
9 * ----------------------------------------------------------
11 * Squid is the result of efforts by numerous individuals from
12 * the Internet community; see the CONTRIBUTORS file for full
13 * details. Many organizations have provided support for Squid's
14 * development; see the SPONSORS file for full details. Squid is
15 * Copyrighted (C) 2001 by the Regents of the University of
16 * California; see the COPYRIGHT file for full details. Squid
17 * incorporates software developed and/or copyrighted by other
18 * sources; see the CREDITS file for full details.
20 * This program is free software; you can redistribute it and/or modify
21 * it under the terms of the GNU General Public License as published by
22 * the Free Software Foundation; either version 2 of the License, or
23 * (at your option) any later version.
25 * This program is distributed in the hope that it will be useful,
26 * but WITHOUT ANY WARRANTY; without even the implied warranty of
27 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
28 * GNU General Public License for more details.
30 * You should have received a copy of the GNU General Public License
31 * along with this program; if not, write to the Free Software
32 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
40 #include "StoreSearch.h"
41 #include "SquidTime.h"
43 static struct _store_rebuild_data counts
;
45 static struct timeval rebuild_start
;
46 static void storeCleanup(void *);
49 /* total number of "swap.state" entries that will be read */
51 /* number of entries read so far */
53 } store_rebuild_progress
;
55 static store_rebuild_progress
*RebuildProgress
= NULL
;
58 storeCleanupDoubleCheck(StoreEntry
* e
)
60 SwapDir
*SD
= dynamic_cast<SwapDir
*>(INDEXSD(e
->swap_dirn
));
61 return (SD
->doubleCheck(*e
));
65 storeCleanup(void *datanotused
)
67 static int store_errors
= 0;
68 static StoreSearchPointer currentSearch
;
69 static int validated
= 0;
71 if (currentSearch
== NULL
|| currentSearch
->isDone())
72 currentSearch
= Store::Root().search(NULL
, NULL
);
74 size_t statCount
= 500;
76 while (statCount
-- && !currentSearch
->isDone() && currentSearch
->next()) {
79 e
= currentSearch
->currentItem();
81 if (EBIT_TEST(e
->flags
, ENTRY_VALIDATED
))
85 * Calling StoreEntry->release() has no effect because we're
86 * still in 'store_rebuilding' state
88 if (e
->swap_filen
< 0)
91 if (opt_store_doublecheck
)
92 if (storeCleanupDoubleCheck(e
))
95 EBIT_SET(e
->flags
, ENTRY_VALIDATED
);
98 * Only set the file bit if we know its a valid entry
99 * otherwise, set it in the validation procedure
103 if (e
->swap_status
== SWAPOUT_DONE
)
104 e
->store()->updateSize(e
->swap_file_sz
, 1);
106 if ((++validated
& 0x3FFFF) == 0)
107 /* TODO format the int with with a stream operator */
108 debugs(20, 1, " " << validated
<< " Entries Validated so far.");
111 if (currentSearch
->isDone()) {
112 debugs(20, 1, " Completed Validation Procedure");
113 debugs(20, 1, " Validated " << validated
<< " Entries");
114 debugs(20, 1, " store_swap_size = " << store_swap_size
);
115 StoreController::store_dirs_rebuilding
--;
116 assert(0 == StoreController::store_dirs_rebuilding
);
118 if (opt_store_doublecheck
)
119 assert(store_errors
== 0);
122 storeDigestNoteStoreReady();
124 currentSearch
= NULL
;
126 eventAdd("storeCleanup", storeCleanup
, NULL
, 0.0, 1);
129 /* meta data recreated from disk image in swap directory */
132 storeRebuildComplete(struct _store_rebuild_data
*dc
)
135 counts
.objcount
+= dc
->objcount
;
136 counts
.expcount
+= dc
->expcount
;
137 counts
.scancount
+= dc
->scancount
;
138 counts
.clashcount
+= dc
->clashcount
;
139 counts
.dupcount
+= dc
->dupcount
;
140 counts
.cancelcount
+= dc
->cancelcount
;
141 counts
.invalid
+= dc
->invalid
;
142 counts
.badflags
+= dc
->badflags
;
143 counts
.bad_log_op
+= dc
->bad_log_op
;
144 counts
.zero_object_sz
+= dc
->zero_object_sz
;
146 * When store_dirs_rebuilding == 1, it means we are done reading
147 * or scanning all cache_dirs. Now report the stats and start
148 * the validation (storeCleanup()) thread.
151 if (StoreController::store_dirs_rebuilding
> 1)
154 dt
= tvSubDsec(rebuild_start
, current_time
);
156 debugs(20, 1, "Finished rebuilding storage from disk.");
157 debugs(20, 1, " " << std::setw(7) << counts
.scancount
<< " Entries scanned");
158 debugs(20, 1, " " << std::setw(7) << counts
.invalid
<< " Invalid entries.");
159 debugs(20, 1, " " << std::setw(7) << counts
.badflags
<< " With invalid flags.");
160 debugs(20, 1, " " << std::setw(7) << counts
.objcount
<< " Objects loaded.");
161 debugs(20, 1, " " << std::setw(7) << counts
.expcount
<< " Objects expired.");
162 debugs(20, 1, " " << std::setw(7) << counts
.cancelcount
<< " Objects cancelled.");
163 debugs(20, 1, " " << std::setw(7) << counts
.dupcount
<< " Duplicate URLs purged.");
164 debugs(20, 1, " " << std::setw(7) << counts
.clashcount
<< " Swapfile clashes avoided.");
165 debugs(20, 1, " Took "<< std::setw(3)<< std::setprecision(2) << dt
<< " seconds ("<< std::setw(6) <<
166 ((double) counts
.objcount
/ (dt
> 0.0 ? dt
: 1.0)) << " objects/sec).");
167 debugs(20, 1, "Beginning Validation Procedure");
169 eventAdd("storeCleanup", storeCleanup
, NULL
, 0.0, 1);
171 xfree(RebuildProgress
);
173 RebuildProgress
= NULL
;
177 * this is ugly. We don't actually start any rebuild threads here,
178 * but only initialize counters, etc. The rebuild threads are
179 * actually started by the filesystem "fooDirInit" function.
182 storeRebuildStart(void)
184 memset(&counts
, '\0', sizeof(counts
));
185 rebuild_start
= current_time
;
187 * Note: store_dirs_rebuilding is initialized to 1.
189 * When we parse the configuration and construct each swap dir,
190 * the construction of that raises the rebuild count.
192 * This prevents us from trying to write clean logs until we
193 * finished rebuilding - including after a reconfiguration that opens an
194 * existing swapdir. The corresponding decrement * occurs in
195 * storeCleanup(), when it is finished.
197 RebuildProgress
= (store_rebuild_progress
*)xcalloc(Config
.cacheSwap
.n_configured
,
198 sizeof(store_rebuild_progress
));
202 * A fs-specific rebuild procedure periodically reports its
206 storeRebuildProgress(int sd_index
, int total
, int sofar
)
208 static time_t last_report
= 0;
215 if (sd_index
>= Config
.cacheSwap
.n_configured
)
218 if (NULL
== RebuildProgress
)
221 RebuildProgress
[sd_index
].total
= total
;
223 RebuildProgress
[sd_index
].scanned
= sofar
;
225 if (squid_curtime
- last_report
< 15)
228 for (sd_index
= 0; sd_index
< Config
.cacheSwap
.n_configured
; sd_index
++) {
229 n
+= (double) RebuildProgress
[sd_index
].scanned
;
230 d
+= (double) RebuildProgress
[sd_index
].total
;
233 debugs(20, 1, "Store rebuilding is "<< std::setw(4)<< std::setprecision(2) << 100.0 * n
/ d
<< "% complete");
234 last_report
= squid_curtime
;
238 #include "StoreMetaUnpacker.h"
239 #include "StoreMeta.h"
242 struct InitStoreEntry
: public unary_function
<StoreMeta
, void> {
243 InitStoreEntry(StoreEntry
*anEntry
, cache_key
*aKey
):what(anEntry
),index(aKey
) {}
245 void operator()(StoreMeta
const &x
) {
246 switch (x
.getType()) {
249 assert(x
.length
== SQUID_MD5_DIGEST_LENGTH
);
250 memcpy(index
, x
.value
, SQUID_MD5_DIGEST_LENGTH
);
263 tmp
= (struct old_metahdr
*)x
.value
;
264 assert(x
.length
== STORE_HDR_METASIZE_OLD
);
265 what
->timestamp
= tmp
->timestamp
;
266 what
->lastref
= tmp
->lastref
;
267 what
->expires
= tmp
->expires
;
268 what
->lastmod
= tmp
->lastmod
;
269 what
->swap_file_sz
= tmp
->swap_file_sz
;
270 what
->refcount
= tmp
->refcount
;
271 what
->flags
= tmp
->flags
;
274 case STORE_META_STD_LFS
:
275 assert(x
.length
== STORE_HDR_METASIZE
);
276 memcpy(&what
->timestamp
, x
.value
, STORE_HDR_METASIZE
);
289 storeRebuildLoadEntry(int fd
, int diskIndex
, MemBuf
&buf
,
290 struct _store_rebuild_data
&counts
)
295 assert(buf
.hasSpace()); // caller must allocate
297 const int len
= FD_READ_METHOD(fd
, buf
.space(), buf
.spaceSize());
298 statCounter
.syscalls
.disk
.reads
++;
300 const int xerrno
= errno
;
301 debugs(47, 1, "cache_dir[" << diskIndex
<< "]: " <<
302 "failed to read swap entry meta data: " << xstrerr(xerrno
));
311 storeRebuildParseEntry(MemBuf
&buf
, StoreEntry
&tmpe
, cache_key
*key
,
312 struct _store_rebuild_data
&counts
,
313 uint64_t expectedSize
)
315 int swap_hdr_len
= 0;
316 StoreMetaUnpacker
aBuilder(buf
.content(), buf
.contentSize(), &swap_hdr_len
);
317 if (aBuilder
.isBufferZero()) {
318 debugs(47,5, HERE
<< "skipping empty record.");
322 if (!aBuilder
.isBufferSane()) {
323 debugs(47,1, "Warning: Ignoring malformed cache entry.");
327 StoreMeta
*tlv_list
= aBuilder
.createStoreMeta();
329 debugs(47, 1, HERE
<< "failed to get swap entry meta data list");
333 // TODO: consume parsed metadata?
335 debugs(47,7, HERE
<< "successful swap meta unpacking");
336 memset(key
, '\0', SQUID_MD5_DIGEST_LENGTH
);
338 InitStoreEntry
visitor(&tmpe
, key
);
339 for_each(*tlv_list
, visitor
);
340 storeSwapTLVFree(tlv_list
);
343 if (storeKeyNull(key
)) {
344 debugs(47,1, HERE
<< "NULL swap entry key");
351 if (expectedSize
> 0) {
352 if (tmpe
.swap_file_sz
== 0) {
353 tmpe
.swap_file_sz
= expectedSize
;
354 } else if (tmpe
.swap_file_sz
== (uint64_t)(expectedSize
- swap_hdr_len
)) {
355 tmpe
.swap_file_sz
= expectedSize
;
356 } else if (tmpe
.swap_file_sz
!= expectedSize
) {
357 debugs(47, 1, HERE
<< "swap entry SIZE MISMATCH " <<
358 tmpe
.swap_file_sz
<< "!=" << expectedSize
);
362 if (tmpe
.swap_file_sz
<= 0) {
363 debugs(47, 1, HERE
<< "missing swap entry size: " << tmpe
);
367 if (EBIT_TEST(tmpe
.flags
, KEY_PRIVATE
)) {
376 storeRebuildKeepEntry(const StoreEntry
&tmpe
, const cache_key
*key
,
377 struct _store_rebuild_data
&counts
)
379 /* this needs to become
381 * 2) make synthetic request with headers ?? or otherwise search
382 * for a matching object in the store
383 * TODO FIXME change to new async api
384 * TODO FIXME I think there is a race condition here with the
386 * store A reads in object foo, searchs for it, and finds nothing.
387 * store B reads in object foo, searchs for it, finds nothing.
388 * store A gets called back with nothing, so registers the object
389 * store B gets called back with nothing, so registers the object,
390 * which will conflict when the in core index gets around to scanning
393 * this suggests that rather than searching for duplicates, the
394 * index rebuild should just assume its the most recent accurate
395 * store entry and whoever indexes the stores handles duplicates.
397 if (StoreEntry
*e
= Store::Root().get(key
)) {
399 if (e
->lastref
>= tmpe
.lastref
) {
400 /* key already exists, old entry is newer */
401 /* keep old, ignore new */
404 // For some stores, get() creates/unpacks a store entry. Signal
405 // such stores that we will no longer use the get() result:
411 /* URL already exists, this swapfile not being used */
412 /* junk old, load new */
413 e
->release(); /* release old entry */