1 |
=head1 NAME |
2 |
|
3 |
BDB - Asynchronous Berkeley DB access |
4 |
|
5 |
=head1 SYNOPSIS |
6 |
|
7 |
use BDB; |
8 |
|
9 |
=head1 DESCRIPTION |
10 |
|
11 |
See the BerkeleyDB documentation (L<http://www.oracle.com/technology/documentation/berkeley-db/db/index.html>). |
12 |
The BDB API is very similar to the C API (the translation has been very faithful). |
13 |
|
14 |
See also the example sections in the document below and possibly the eg/ |
15 |
subdirectory of the BDB distribution. Last not least see the IO::AIO |
16 |
documentation, as that module uses almost the same asynchronous request |
17 |
model as this module. |
18 |
|
19 |
I know this is woefully inadequate documentation. Send a patch! |
20 |
|
21 |
|
22 |
=head1 REQUEST ANATOMY AND LIFETIME |
23 |
|
24 |
Every request method creates a request. which is a C data structure not |
25 |
directly visible to Perl. |
26 |
|
27 |
During their existance, bdb requests travel through the following states, |
28 |
in order: |
29 |
|
30 |
=over 4 |
31 |
|
32 |
=item ready |
33 |
|
34 |
Immediately after a request is created it is put into the ready state, |
35 |
waiting for a thread to execute it. |
36 |
|
37 |
=item execute |
38 |
|
39 |
A thread has accepted the request for processing and is currently |
40 |
executing it (e.g. blocking in read). |
41 |
|
42 |
=item pending |
43 |
|
44 |
The request has been executed and is waiting for result processing. |
45 |
|
46 |
While request submission and execution is fully asynchronous, result |
47 |
processing is not and relies on the perl interpreter calling C<poll_cb> |
48 |
(or another function with the same effect). |
49 |
|
50 |
=item result |
51 |
|
52 |
The request results are processed synchronously by C<poll_cb>. |
53 |
|
54 |
The C<poll_cb> function will process all outstanding aio requests by |
55 |
calling their callbacks, freeing memory associated with them and managing |
56 |
any groups they are contained in. |
57 |
|
58 |
=item done |
59 |
|
60 |
Request has reached the end of its lifetime and holds no resources anymore |
61 |
(except possibly for the Perl object, but its connection to the actual |
62 |
aio request is severed and calling its methods will either do nothing or |
63 |
result in a runtime error). |
64 |
|
65 |
=back |
66 |
|
67 |
=cut |
68 |
|
69 |
package BDB; |
70 |
|
71 |
no warnings; |
72 |
use strict 'vars'; |
73 |
|
74 |
use base 'Exporter'; |
75 |
|
76 |
BEGIN { |
77 |
our $VERSION = '1.1'; |
78 |
|
79 |
our @BDB_REQ = qw( |
80 |
db_env_open db_env_close db_env_txn_checkpoint db_env_lock_detect |
81 |
db_env_memp_sync db_env_memp_trickle |
82 |
db_open db_close db_compact db_sync db_put db_get db_pget db_del db_key_range |
83 |
db_txn_commit db_txn_abort db_txn_finish |
84 |
db_c_close db_c_count db_c_put db_c_get db_c_pget db_c_del |
85 |
db_sequence_open db_sequence_close |
86 |
db_sequence_get db_sequence_remove |
87 |
); |
88 |
our @EXPORT = (@BDB_REQ, qw(dbreq_pri dbreq_nice db_env_create db_create)); |
89 |
our @EXPORT_OK = qw( |
90 |
poll_fileno poll_cb poll_wait flush |
91 |
min_parallel max_parallel max_idle |
92 |
nreqs nready npending nthreads |
93 |
max_poll_time max_poll_reqs |
94 |
); |
95 |
|
96 |
require XSLoader; |
97 |
XSLoader::load ("BDB", $VERSION); |
98 |
} |
99 |
|
100 |
=head2 BERKELEYDB FUNCTIONS |
101 |
|
102 |
All of these are functions. The create functions simply return a new |
103 |
object and never block. All the remaining functions all take an optional |
104 |
callback as last argument. If it is missing, then the fucntion will be |
105 |
executed synchronously. |
106 |
|
107 |
BDB functions that cannot block (mostly functions that manipulate |
108 |
settings) are method calls on the relevant objects, so the rule of thumb |
109 |
is: if its a method, its not blocking, if its a function, it takes a |
110 |
callback as last argument. |
111 |
|
112 |
In the following, C<$int> signifies an integer return value, |
113 |
C<octetstring> is a "binary string" (i.e. a perl string with no character |
114 |
indices >255), C<U32> is an unsigned 32 bit integer, C<int> is some |
115 |
integer, C<NV> is a floating point value. |
116 |
|
117 |
The C<SV *> types are generic perl scalars (for input and output of data |
118 |
values), and the C<SV *callback> is the optional callback function to call |
119 |
when the request is completed. |
120 |
|
121 |
The various C<DB_ENV> etc. arguments are handles return by |
122 |
C<db_env_create>, C<db_create>, C<txn_begin> and so on. If they have an |
123 |
appended C<_ornull> this means they are optional and you can pass C<undef> |
124 |
for them, resulting a NULL pointer on the C level. |
125 |
|
126 |
=head3 BDB functions |
127 |
|
128 |
Functions in the BDB namespace, exported by default: |
129 |
|
130 |
$env = db_env_create (U32 env_flags = 0) |
131 |
flags: RPCCLIENT |
132 |
|
133 |
db_env_open (DB_ENV *env, octetstring db_home, U32 open_flags, int mode, SV *callback = &PL_sv_undef) |
134 |
open_flags: INIT_CDB INIT_LOCK INIT_LOG INIT_MPOOL INIT_REP INIT_TXN RECOVER RECOVER_FATAL USE_ENVIRON USE_ENVIRON_ROOT CREATE LOCKDOWN PRIVATE REGISTER SYSTEM_MEM |
135 |
db_env_close (DB_ENV *env, U32 flags = 0, SV *callback = &PL_sv_undef) |
136 |
db_env_txn_checkpoint (DB_ENV *env, U32 kbyte = 0, U32 min = 0, U32 flags = 0, SV *callback = &PL_sv_undef) |
137 |
flags: FORCE |
138 |
db_env_lock_detect (DB_ENV *env, U32 flags = 0, U32 atype = DB_LOCK_DEFAULT, SV *dummy = 0, SV *callback = &PL_sv_undef) |
139 |
atype: LOCK_DEFAULT LOCK_EXPIRE LOCK_MAXLOCKS LOCK_MAXWRITE LOCK_MINLOCKS LOCK_MINWRITE LOCK_OLDEST LOCK_RANDOM LOCK_YOUNGEST |
140 |
db_env_memp_sync (DB_ENV *env, SV *dummy = 0, SV *callback = &PL_sv_undef) |
141 |
db_env_memp_trickle (DB_ENV *env, int percent, SV *dummy = 0, SV *callback = &PL_sv_undef) |
142 |
|
143 |
$db = db_create (DB_ENV *env = 0, U32 flags = 0) |
144 |
flags: XA_CREATE |
145 |
|
146 |
db_open (DB *db, DB_TXN_ornull *txnid, octetstring file, octetstring database, int type, U32 flags, int mode, SV *callback = &PL_sv_undef) |
147 |
flags: AUTO_COMMIT CREATE EXCL MULTIVERSION NOMMAP RDONLY READ_UNCOMMITTED THREAD TRUNCATE |
148 |
db_close (DB *db, U32 flags = 0, SV *callback = &PL_sv_undef) |
149 |
flags: DB_NOSYNC |
150 |
db_compact (DB *db, DB_TXN_ornull *txn = 0, SV *start = 0, SV *stop = 0, SV *unused1 = 0, U32 flags = DB_FREE_SPACE, SV *unused2 = 0, SV *callback = &PL_sv_undef) |
151 |
flags: FREELIST_ONLY FREE_SPACE |
152 |
db_sync (DB *db, U32 flags = 0, SV *callback = &PL_sv_undef) |
153 |
db_key_range (DB *db, DB_TXN_ornull *txn, SV *key, SV *key_range, U32 flags = 0, SV *callback = &PL_sv_undef) |
154 |
db_put (DB *db, DB_TXN_ornull *txn, SV *key, SV *data, U32 flags = 0, SV *callback = &PL_sv_undef) |
155 |
flags: APPEND NODUPDATA NOOVERWRITE |
156 |
db_get (DB *db, DB_TXN_ornull *txn, SV *key, SV *data, U32 flags = 0, SV *callback = &PL_sv_undef) |
157 |
flags: CONSUME CONSUME_WAIT GET_BOTH SET_RECNO MULTIPLE READ_COMMITTED READ_UNCOMMITTED RMW |
158 |
db_pget (DB *db, DB_TXN_ornull *txn, SV *key, SV *pkey, SV *data, U32 flags = 0, SV *callback = &PL_sv_undef) |
159 |
flags: CONSUME CONSUME_WAIT GET_BOTH SET_RECNO MULTIPLE READ_COMMITTED READ_UNCOMMITTED RMW |
160 |
db_del (DB *db, DB_TXN_ornull *txn, SV *key, U32 flags = 0, SV *callback = &PL_sv_undef) |
161 |
db_txn_commit (DB_TXN *txn, U32 flags = 0, SV *callback = &PL_sv_undef) |
162 |
flags: TXN_NOSYNC TXN_SYNC |
163 |
db_txn_abort (DB_TXN *txn, SV *callback = &PL_sv_undef) |
164 |
|
165 |
db_c_close (DBC *dbc, SV *callback = &PL_sv_undef) |
166 |
db_c_count (DBC *dbc, SV *count, U32 flags = 0, SV *callback = &PL_sv_undef) |
167 |
db_c_put (DBC *dbc, SV *key, SV *data, U32 flags = 0, SV *callback = &PL_sv_undef) |
168 |
flags: AFTER BEFORE CURRENT KEYFIRST KEYLAST NODUPDATA |
169 |
db_c_get (DBC *dbc, SV *key, SV *data, U32 flags = 0, SV *callback = &PL_sv_undef) |
170 |
flags: CURRENT FIRST GET_BOTH GET_BOTH_RANGE GET_RECNO JOIN_ITEM LAST NEXT NEXT_DUP NEXT_NODUP PREV PREV_DUP PREV_NODUP SET SET_RANGE SET_RECNO READ_UNCOMMITTED MULTIPLE MULTIPLE_KEY RMW |
171 |
db_c_pget (DBC *dbc, SV *key, SV *pkey, SV *data, U32 flags = 0, SV *callback = &PL_sv_undef) |
172 |
db_c_del (DBC *dbc, U32 flags = 0, SV *callback = &PL_sv_undef) |
173 |
|
174 |
db_sequence_open (DB_SEQUENCE *seq, DB_TXN_ornull *txnid, SV *key, U32 flags = 0, SV *callback = &PL_sv_undef) |
175 |
flags: CREATE EXCL |
176 |
db_sequence_close (DB_SEQUENCE *seq, U32 flags = 0, SV *callback = &PL_sv_undef) |
177 |
db_sequence_get (DB_SEQUENCE *seq, DB_TXN_ornull *txnid, int delta, SV *seq_value, U32 flags = DB_TXN_NOSYNC, SV *callback = &PL_sv_undef) |
178 |
flags: TXN_NOSYNC |
179 |
db_sequence_remove (DB_SEQUENCE *seq, DB_TXN_ornull *txnid = 0, U32 flags = 0, SV *callback = &PL_sv_undef) |
180 |
flags: TXN_NOSYNC |
181 |
|
182 |
=head4 db_txn_finish (DB_TXN *txn, U32 flags = 0, SV *callback = &PL_sv_undef) |
183 |
|
184 |
This is not actually a Berkeley DB function but a BDB module |
185 |
extension. The background for this exytension is: It is very annoying to |
186 |
have to check every single BDB function for error returns and provide a |
187 |
codepath out of your transaction. While the BDB module still makes this |
188 |
possible, it contains the following extensions: |
189 |
|
190 |
When a transaction-protected function returns any operating system |
191 |
error (errno > 0), BDB will set the C<TXN_DEADLOCK> flag on the |
192 |
transaction. This flag is also set by Berkeley DB functions themselves |
193 |
when an operation fails with LOCK_DEADLOCK, and it causes all further |
194 |
operations on that transaction (including C<db_txn_commit>) to fail. |
195 |
|
196 |
The C<db_txn_finish> request will look at this flag, and, if it is set, |
197 |
will automatically call C<db_txn_abort> (setting errno to C<LOCK_DEADLOCK> |
198 |
if it isn't set to something else yet). If it isn't set, it will call |
199 |
C<db_txn_commit> and return the error normally. |
200 |
|
201 |
How to use this? Easy: just write your transaction normally: |
202 |
|
203 |
my $txn = $db_env->txn_begin; |
204 |
db_get $db, $txn, "key", my $data; |
205 |
db_put $db, $txn, "key", $data + 1 unless $! == BDB::NOTFOUND; |
206 |
db_txn_finish $txn; |
207 |
die "transaction failed" if $!; |
208 |
|
209 |
That is, handle only the expected errors. If something unexpected happens |
210 |
(EIO, LOCK_NOTGRANTED or a deadlock in either db_get or db_put), then the remaining |
211 |
requests (db_put in this case) will simply be skipped (they will fail with |
212 |
LOCK_DEADLOCK) and the transaction will be aborted. |
213 |
|
214 |
You can use the C<< $txn->failed >> method to check wether a transaction |
215 |
has failed in this way and abort further processing (excluding |
216 |
C<db_txn_finish>). |
217 |
|
218 |
=head3 DB_ENV/database environment methods |
219 |
|
220 |
Methods available on DB_ENV/$env handles: |
221 |
|
222 |
DESTROY (DB_ENV_ornull *env) |
223 |
CODE: |
224 |
if (env) |
225 |
env->close (env, 0); |
226 |
|
227 |
$int = $env->set_data_dir (const char *dir) |
228 |
$int = $env->set_tmp_dir (const char *dir) |
229 |
$int = $env->set_lg_dir (const char *dir) |
230 |
$int = $env->set_shm_key (long shm_key) |
231 |
$int = $env->set_cachesize (U32 gbytes, U32 bytes, int ncache = 0) |
232 |
$int = $env->set_flags (U32 flags, int onoff) |
233 |
$env->set_errfile (FILE *errfile = 0) |
234 |
$env->set_msgfile (FILE *msgfile = 0) |
235 |
$int = $env->set_verbose (U32 which, int onoff = 1) |
236 |
$int = $env->set_encrypt (const char *password, U32 flags = 0) |
237 |
$int = $env->set_timeout (NV timeout_seconds, U32 flags = SET_TXN_TIMEOUT) |
238 |
$int = $env->set_mp_max_openfd (int maxopenfd); |
239 |
$int = $env->set_mp_max_write (int maxwrite, int maxwrite_sleep); |
240 |
$int = $env->set_mp_mmapsize (int mmapsize_mb) |
241 |
$int = $env->set_lk_detect (U32 detect = DB_LOCK_DEFAULT) |
242 |
$int = $env->set_lk_max_lockers (U32 max) |
243 |
$int = $env->set_lk_max_locks (U32 max) |
244 |
$int = $env->set_lk_max_objects (U32 max) |
245 |
$int = $env->set_lg_bsize (U32 max) |
246 |
$int = $env->set_lg_max (U32 max) |
247 |
|
248 |
$txn = $env->txn_begin (DB_TXN_ornull *parent = 0, U32 flags = 0) |
249 |
flags: READ_COMMITTED READ_UNCOMMITTED TXN_NOSYNC TXN_NOWAIT TXN_SNAPSHOT TXN_SYNC TXN_WAIT TXN_WRITE_NOSYNC |
250 |
|
251 |
=head4 Example: |
252 |
|
253 |
use AnyEvent; |
254 |
use BDB; |
255 |
|
256 |
our $FH; open $FH, "<&=" . BDB::poll_fileno; |
257 |
our $WATCHER = AnyEvent->io (fh => $FH, poll => 'r', cb => \&BDB::poll_cb); |
258 |
|
259 |
BDB::min_parallel 8; |
260 |
|
261 |
my $env = db_env_create; |
262 |
|
263 |
mkdir "bdtest", 0700; |
264 |
db_env_open |
265 |
$env, |
266 |
"bdtest", |
267 |
BDB::INIT_LOCK | BDB::INIT_LOG | BDB::INIT_MPOOL | BDB::INIT_TXN | BDB::RECOVER | BDB::USE_ENVIRON | BDB::CREATE, |
268 |
0600; |
269 |
|
270 |
$env->set_flags (BDB::AUTO_COMMIT | BDB::TXN_NOSYNC, 1); |
271 |
|
272 |
|
273 |
=head3 DB/database methods |
274 |
|
275 |
Methods available on DB/$db handles: |
276 |
|
277 |
DESTROY (DB_ornull *db) |
278 |
CODE: |
279 |
if (db) |
280 |
{ |
281 |
SV *env = (SV *)db->app_private; |
282 |
db->close (db, 0); |
283 |
SvREFCNT_dec (env); |
284 |
} |
285 |
|
286 |
$int = $db->set_cachesize (U32 gbytes, U32 bytes, int ncache = 0) |
287 |
$int = $db->set_flags (U32 flags) |
288 |
flags: CHKSUM ENCRYPT TXN_NOT_DURABLE |
289 |
Btree: DUP DUPSORT RECNUM REVSPLITOFF |
290 |
Hash: DUP DUPSORT |
291 |
Queue: INORDER |
292 |
Recno: RENUMBER SNAPSHOT |
293 |
|
294 |
$int = $db->set_encrypt (const char *password, U32 flags) |
295 |
$int = $db->set_lorder (int lorder) |
296 |
$int = $db->set_bt_minkey (U32 minkey) |
297 |
$int = $db->set_re_delim (int delim) |
298 |
$int = $db->set_re_pad (int re_pad) |
299 |
$int = $db->set_re_source (char *source) |
300 |
$int = $db->set_re_len (U32 re_len) |
301 |
$int = $db->set_h_ffactor (U32 h_ffactor) |
302 |
$int = $db->set_h_nelem (U32 h_nelem) |
303 |
$int = $db->set_q_extentsize (U32 extentsize) |
304 |
|
305 |
$dbc = $db->cursor (DB_TXN_ornull *txn = 0, U32 flags = 0) |
306 |
flags: READ_COMMITTED READ_UNCOMMITTED WRITECURSOR TXN_SNAPSHOT |
307 |
$seq = $db->sequence (U32 flags = 0) |
308 |
|
309 |
=head4 Example: |
310 |
|
311 |
my $db = db_create $env; |
312 |
db_open $db, undef, "table", undef, BDB::BTREE, BDB::AUTO_COMMIT | BDB::CREATE | BDB::READ_UNCOMMITTED, 0600; |
313 |
|
314 |
for (1..1000) { |
315 |
db_put $db, undef, "key $_", "data $_"; |
316 |
|
317 |
db_key_range $db, undef, "key $_", my $keyrange; |
318 |
my ($lt, $eq, $gt) = @$keyrange; |
319 |
} |
320 |
|
321 |
db_del $db, undef, "key $_" for 1..1000; |
322 |
|
323 |
db_sync $db; |
324 |
|
325 |
|
326 |
=head3 DB_TXN/transaction methods |
327 |
|
328 |
Methods available on DB_TXN/$txn handles: |
329 |
|
330 |
DESTROY (DB_TXN_ornull *txn) |
331 |
CODE: |
332 |
if (txn) |
333 |
txn->abort (txn); |
334 |
|
335 |
$int = $txn->set_timeout (NV timeout_seconds, U32 flags = SET_TXN_TIMEOUT) |
336 |
flags: SET_LOCK_TIMEOUT SET_TXN_TIMEOUT |
337 |
|
338 |
$bool = $txn->failed |
339 |
# see db_txn_finish documentation, above |
340 |
|
341 |
|
342 |
=head3 DBC/cursor methods |
343 |
|
344 |
Methods available on DBC/$dbc handles: |
345 |
|
346 |
DESTROY (DBC_ornull *dbc) |
347 |
CODE: |
348 |
if (dbc) |
349 |
dbc->c_close (dbc); |
350 |
|
351 |
=head4 Example: |
352 |
|
353 |
my $c = $db->cursor; |
354 |
|
355 |
for (;;) { |
356 |
db_c_get $c, my $key, my $data, BDB::NEXT; |
357 |
warn "<$!,$key,$data>"; |
358 |
last if $!; |
359 |
} |
360 |
|
361 |
db_c_close $c; |
362 |
|
363 |
|
364 |
=head3 DB_SEQUENCE/sequence methods |
365 |
|
366 |
Methods available on DB_SEQUENCE/$seq handles: |
367 |
|
368 |
DESTROY (DB_SEQUENCE_ornull *seq) |
369 |
CODE: |
370 |
if (seq) |
371 |
seq->close (seq, 0); |
372 |
|
373 |
$int = $seq->initial_value (db_seq_t value) |
374 |
$int = $seq->set_cachesize (U32 size) |
375 |
$int = $seq->set_flags (U32 flags) |
376 |
flags: SEQ_DEC SEQ_INC SEQ_WRAP |
377 |
$int = $seq->set_range (db_seq_t min, db_seq_t max) |
378 |
|
379 |
=head4 Example: |
380 |
|
381 |
my $seq = $db->sequence; |
382 |
|
383 |
db_sequence_open $seq, undef, "seq", BDB::CREATE; |
384 |
db_sequence_get $seq, undef, 1, my $value; |
385 |
|
386 |
|
387 |
=head2 SUPPORT FUNCTIONS |
388 |
|
389 |
=head3 EVENT PROCESSING AND EVENT LOOP INTEGRATION |
390 |
|
391 |
=over 4 |
392 |
|
393 |
=item $fileno = BDB::poll_fileno |
394 |
|
395 |
Return the I<request result pipe file descriptor>. This filehandle must be |
396 |
polled for reading by some mechanism outside this module (e.g. Event or |
397 |
select, see below or the SYNOPSIS). If the pipe becomes readable you have |
398 |
to call C<poll_cb> to check the results. |
399 |
|
400 |
See C<poll_cb> for an example. |
401 |
|
402 |
=item BDB::poll_cb |
403 |
|
404 |
Process some outstanding events on the result pipe. You have to call this |
405 |
regularly. Returns the number of events processed. Returns immediately |
406 |
when no events are outstanding. The amount of events processed depends on |
407 |
the settings of C<BDB::max_poll_req> and C<BDB::max_poll_time>. |
408 |
|
409 |
If not all requests were processed for whatever reason, the filehandle |
410 |
will still be ready when C<poll_cb> returns. |
411 |
|
412 |
Example: Install an Event watcher that automatically calls |
413 |
BDB::poll_cb with high priority: |
414 |
|
415 |
Event->io (fd => BDB::poll_fileno, |
416 |
poll => 'r', async => 1, |
417 |
cb => \&BDB::poll_cb); |
418 |
|
419 |
=item BDB::max_poll_reqs $nreqs |
420 |
|
421 |
=item BDB::max_poll_time $seconds |
422 |
|
423 |
These set the maximum number of requests (default C<0>, meaning infinity) |
424 |
that are being processed by C<BDB::poll_cb> in one call, respectively |
425 |
the maximum amount of time (default C<0>, meaning infinity) spent in |
426 |
C<BDB::poll_cb> to process requests (more correctly the mininum amount |
427 |
of time C<poll_cb> is allowed to use). |
428 |
|
429 |
Setting C<max_poll_time> to a non-zero value creates an overhead of one |
430 |
syscall per request processed, which is not normally a problem unless your |
431 |
callbacks are really really fast or your OS is really really slow (I am |
432 |
not mentioning Solaris here). Using C<max_poll_reqs> incurs no overhead. |
433 |
|
434 |
Setting these is useful if you want to ensure some level of |
435 |
interactiveness when perl is not fast enough to process all requests in |
436 |
time. |
437 |
|
438 |
For interactive programs, values such as C<0.01> to C<0.1> should be fine. |
439 |
|
440 |
Example: Install an Event watcher that automatically calls |
441 |
BDB::poll_cb with low priority, to ensure that other parts of the |
442 |
program get the CPU sometimes even under high AIO load. |
443 |
|
444 |
# try not to spend much more than 0.1s in poll_cb |
445 |
BDB::max_poll_time 0.1; |
446 |
|
447 |
# use a low priority so other tasks have priority |
448 |
Event->io (fd => BDB::poll_fileno, |
449 |
poll => 'r', nice => 1, |
450 |
cb => &BDB::poll_cb); |
451 |
|
452 |
=item BDB::poll_wait |
453 |
|
454 |
If there are any outstanding requests and none of them in the result |
455 |
phase, wait till the result filehandle becomes ready for reading (simply |
456 |
does a C<select> on the filehandle. This is useful if you want to |
457 |
synchronously wait for some requests to finish). |
458 |
|
459 |
See C<nreqs> for an example. |
460 |
|
461 |
=item BDB::poll |
462 |
|
463 |
Waits until some requests have been handled. |
464 |
|
465 |
Returns the number of requests processed, but is otherwise strictly |
466 |
equivalent to: |
467 |
|
468 |
BDB::poll_wait, BDB::poll_cb |
469 |
|
470 |
=item BDB::flush |
471 |
|
472 |
Wait till all outstanding AIO requests have been handled. |
473 |
|
474 |
Strictly equivalent to: |
475 |
|
476 |
BDB::poll_wait, BDB::poll_cb |
477 |
while BDB::nreqs; |
478 |
|
479 |
=back |
480 |
|
481 |
=head3 CONTROLLING THE NUMBER OF THREADS |
482 |
|
483 |
=over 4 |
484 |
|
485 |
=item BDB::min_parallel $nthreads |
486 |
|
487 |
Set the minimum number of AIO threads to C<$nthreads>. The current |
488 |
default is C<8>, which means eight asynchronous operations can execute |
489 |
concurrently at any one time (the number of outstanding requests, |
490 |
however, is unlimited). |
491 |
|
492 |
BDB starts threads only on demand, when an AIO request is queued and |
493 |
no free thread exists. Please note that queueing up a hundred requests can |
494 |
create demand for a hundred threads, even if it turns out that everything |
495 |
is in the cache and could have been processed faster by a single thread. |
496 |
|
497 |
It is recommended to keep the number of threads relatively low, as some |
498 |
Linux kernel versions will scale negatively with the number of threads |
499 |
(higher parallelity => MUCH higher latency). With current Linux 2.6 |
500 |
versions, 4-32 threads should be fine. |
501 |
|
502 |
Under most circumstances you don't need to call this function, as the |
503 |
module selects a default that is suitable for low to moderate load. |
504 |
|
505 |
=item BDB::max_parallel $nthreads |
506 |
|
507 |
Sets the maximum number of AIO threads to C<$nthreads>. If more than the |
508 |
specified number of threads are currently running, this function kills |
509 |
them. This function blocks until the limit is reached. |
510 |
|
511 |
While C<$nthreads> are zero, aio requests get queued but not executed |
512 |
until the number of threads has been increased again. |
513 |
|
514 |
This module automatically runs C<max_parallel 0> at program end, to ensure |
515 |
that all threads are killed and that there are no outstanding requests. |
516 |
|
517 |
Under normal circumstances you don't need to call this function. |
518 |
|
519 |
=item BDB::max_idle $nthreads |
520 |
|
521 |
Limit the number of threads (default: 4) that are allowed to idle (i.e., |
522 |
threads that did not get a request to process within 10 seconds). That |
523 |
means if a thread becomes idle while C<$nthreads> other threads are also |
524 |
idle, it will free its resources and exit. |
525 |
|
526 |
This is useful when you allow a large number of threads (e.g. 100 or 1000) |
527 |
to allow for extremely high load situations, but want to free resources |
528 |
under normal circumstances (1000 threads can easily consume 30MB of RAM). |
529 |
|
530 |
The default is probably ok in most situations, especially if thread |
531 |
creation is fast. If thread creation is very slow on your system you might |
532 |
want to use larger values. |
533 |
|
534 |
=item $oldmaxreqs = BDB::max_outstanding $maxreqs |
535 |
|
536 |
This is a very bad function to use in interactive programs because it |
537 |
blocks, and a bad way to reduce concurrency because it is inexact: Better |
538 |
use an C<aio_group> together with a feed callback. |
539 |
|
540 |
Sets the maximum number of outstanding requests to C<$nreqs>. If you |
541 |
to queue up more than this number of requests, the next call to the |
542 |
C<poll_cb> (and C<poll_some> and other functions calling C<poll_cb>) |
543 |
function will block until the limit is no longer exceeded. |
544 |
|
545 |
The default value is very large, so there is no practical limit on the |
546 |
number of outstanding requests. |
547 |
|
548 |
You can still queue as many requests as you want. Therefore, |
549 |
C<max_oustsanding> is mainly useful in simple scripts (with low values) or |
550 |
as a stop gap to shield against fatal memory overflow (with large values). |
551 |
|
552 |
=item BDB::set_sync_prepare $cb |
553 |
|
554 |
Sets a callback that is called whenever a request is created without an |
555 |
explicit callback. It has to return two code references. The first is used |
556 |
as the request callback, and the second is called to wait until the first |
557 |
callback has been called. The default implementation works like this: |
558 |
|
559 |
sub { |
560 |
my $status; |
561 |
( |
562 |
sub { $status = $! }, |
563 |
sub { BDB::poll while !defined $status; $! = $status }, |
564 |
) |
565 |
} |
566 |
|
567 |
=back |
568 |
|
569 |
=head3 STATISTICAL INFORMATION |
570 |
|
571 |
=over 4 |
572 |
|
573 |
=item BDB::nreqs |
574 |
|
575 |
Returns the number of requests currently in the ready, execute or pending |
576 |
states (i.e. for which their callback has not been invoked yet). |
577 |
|
578 |
Example: wait till there are no outstanding requests anymore: |
579 |
|
580 |
BDB::poll_wait, BDB::poll_cb |
581 |
while BDB::nreqs; |
582 |
|
583 |
=item BDB::nready |
584 |
|
585 |
Returns the number of requests currently in the ready state (not yet |
586 |
executed). |
587 |
|
588 |
=item BDB::npending |
589 |
|
590 |
Returns the number of requests currently in the pending state (executed, |
591 |
but not yet processed by poll_cb). |
592 |
|
593 |
=back |
594 |
|
595 |
=cut |
596 |
|
597 |
set_sync_prepare { |
598 |
my $status; |
599 |
( |
600 |
sub { |
601 |
$status = $!; |
602 |
}, |
603 |
sub { |
604 |
BDB::poll while !defined $status; |
605 |
$! = $status; |
606 |
}, |
607 |
) |
608 |
}; |
609 |
|
610 |
min_parallel 8; |
611 |
|
612 |
END { flush } |
613 |
|
614 |
1; |
615 |
|
616 |
=head2 FORK BEHAVIOUR |
617 |
|
618 |
This module should do "the right thing" when the process using it forks: |
619 |
|
620 |
Before the fork, IO::AIO enters a quiescent state where no requests |
621 |
can be added in other threads and no results will be processed. After |
622 |
the fork the parent simply leaves the quiescent state and continues |
623 |
request/result processing, while the child frees the request/result queue |
624 |
(so that the requests started before the fork will only be handled in the |
625 |
parent). Threads will be started on demand until the limit set in the |
626 |
parent process has been reached again. |
627 |
|
628 |
In short: the parent will, after a short pause, continue as if fork had |
629 |
not been called, while the child will act as if IO::AIO has not been used |
630 |
yet. |
631 |
|
632 |
=head2 MEMORY USAGE |
633 |
|
634 |
Per-request usage: |
635 |
|
636 |
Each aio request uses - depending on your architecture - around 100-200 |
637 |
bytes of memory. In addition, stat requests need a stat buffer (possibly |
638 |
a few hundred bytes), readdir requires a result buffer and so on. Perl |
639 |
scalars and other data passed into aio requests will also be locked and |
640 |
will consume memory till the request has entered the done state. |
641 |
|
642 |
This is not awfully much, so queuing lots of requests is not usually a |
643 |
problem. |
644 |
|
645 |
Per-thread usage: |
646 |
|
647 |
In the execution phase, some aio requests require more memory for |
648 |
temporary buffers, and each thread requires a stack and other data |
649 |
structures (usually around 16k-128k, depending on the OS). |
650 |
|
651 |
=head1 KNOWN BUGS |
652 |
|
653 |
Known bugs will be fixed in the next release, except: |
654 |
|
655 |
If you use a transaction in any request, and the request returns |
656 |
with an operating system error or DB_LOCK_NOTGRANTED, the internal |
657 |
TXN_DEADLOCK flag will be set on the transaction. See C<db_txn_finish>, |
658 |
above. |
659 |
|
660 |
=head1 SEE ALSO |
661 |
|
662 |
L<Coro::AIO>. |
663 |
|
664 |
=head1 AUTHOR |
665 |
|
666 |
Marc Lehmann <schmorp@schmorp.de> |
667 |
http://home.schmorp.de/ |
668 |
|
669 |
=cut |
670 |
|