Bias up recently added tracks. See new_bias and new_bias_age in
[disorder] / server / choose.c
CommitLineData
bea6f6d5
RK
1/*
2 * This file is part of DisOrder
3 * Copyright (C) 2008 Richard Kettlewell
fd9000bc 4 * Copyright (C) 2008 Mark Wooding
bea6f6d5
RK
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
19 * USA
20 */
21/** @file choose.c
22 * @brief Random track chooser
23 *
24 * Picks a track at random and writes it to standard output. If for
25 * any reason no track can be picked - even a trivial reason like a
26 * deadlock - it just exits and expects the server to try again.
27 */
28
29#include <config.h>
30#include "types.h"
31
32#include <getopt.h>
33#include <stdio.h>
34#include <stdlib.h>
35#include <db.h>
36#include <locale.h>
37#include <errno.h>
38#include <sys/types.h>
39#include <unistd.h>
40#include <pcre.h>
41#include <string.h>
42#include <fcntl.h>
43#include <syslog.h>
91c9324a 44#include <time.h>
bea6f6d5
RK
45
46#include "configuration.h"
47#include "log.h"
48#include "defs.h"
49#include "mem.h"
50#include "kvp.h"
51#include "syscalls.h"
52#include "printf.h"
53#include "trackdb.h"
54#include "trackdb-int.h"
55#include "version.h"
91c9324a 56#include "trackname.h"
28dc2d22
RK
57#include "queue.h"
58#include "server-queue.h"
bea6f6d5 59
05dcfac6
RK
60#define BASE_WEIGHT 90000
61
bea6f6d5
RK
62static DB_TXN *global_tid;
63
64static const struct option options[] = {
65 { "help", no_argument, 0, 'h' },
66 { "version", no_argument, 0, 'V' },
67 { "config", required_argument, 0, 'c' },
68 { "debug", no_argument, 0, 'd' },
69 { "no-debug", no_argument, 0, 'D' },
70 { "syslog", no_argument, 0, 's' },
71 { "no-syslog", no_argument, 0, 'S' },
72 { 0, 0, 0, 0 }
73};
74
75/* display usage message and terminate */
76static void help(void) {
77 xprintf("Usage:\n"
78 " disorder-choose [OPTIONS]\n"
79 "Options:\n"
80 " --help, -h Display usage message\n"
81 " --version, -V Display version number\n"
82 " --config PATH, -c PATH Set configuration file\n"
83 " --debug, -d Turn on debugging\n"
84 " --[no-]syslog Enable/disable logging to syslog\n"
85 "\n"
86 "Track choose for DisOrder. Not intended to be run\n"
87 "directly.\n");
88 xfclose(stdout);
89 exit(0);
90}
bea6f6d5
RK
91/** @brief Sum of all weights */
92static unsigned long long total_weight;
93
74b1f70d 94/** @brief The winning track */
95static const char *winning = 0;
96
bea6f6d5
RK
97/** @brief Count of tracks */
98static long ntracks;
99
91c9324a
RK
100static char **required_tags;
101static char **prohibited_tags;
102
28dc2d22
RK
103static int queue_contains(const struct queue_entry *head,
104 const char *track) {
105 const struct queue_entry *q;
106
107 for(q = head->next; q != head; q = q->next)
108 if(!strcmp(q->track, track))
109 return 1;
110 return 0;
111}
112
bea6f6d5
RK
113/** @brief Compute the weight of a track
114 * @param track Track name (UTF-8)
115 * @param data Track data
116 * @param prefs Track preferences
117 * @return Track weight (non-negative)
118 *
119 * Tracks to be excluded entirely are given a weight of 0.
120 */
91c9324a
RK
121static unsigned long compute_weight(const char *track,
122 struct kvp *data,
bea6f6d5
RK
123 struct kvp *prefs) {
124 const char *s;
91c9324a 125 char **track_tags;
05dcfac6 126 time_t last, now = time(0);
91c9324a
RK
127
128 /* Reject tracks not in any collection (race between edit config and
129 * rescan) */
130 if(!find_track_root(track)) {
131 info("found track not in any collection: %s", track);
132 return 0;
133 }
bea6f6d5 134
91c9324a
RK
135 /* Reject aliases to avoid giving aliased tracks extra weight */
136 if(kvp_get(data, "_alias_for"))
137 return 0;
138
139 /* Reject tracks with random play disabled */
bea6f6d5
RK
140 if((s = kvp_get(prefs, "pick_at_random"))
141 && !strcmp(s, "0"))
142 return 0;
91c9324a
RK
143
144 /* Reject tracks played within the last 8 hours */
145 if((s = kvp_get(prefs, "played_time"))) {
146 last = atoll(s);
cebe3127 147 if(now < last + config->replay_min)
91c9324a
RK
148 return 0;
149 }
150
28dc2d22
RK
151 /* Reject tracks currently in the queue or in the recent list */
152 if(queue_contains(&qhead, track)
153 || queue_contains(&phead, track))
154 return 0;
155
91c9324a
RK
156 /* We'll need tags for a number of things */
157 track_tags = parsetags(kvp_get(prefs, "tags"));
158
159 /* Reject tracks with prohibited tags */
160 if(prohibited_tags && tag_intersection(track_tags, prohibited_tags))
161 return 0;
162
163 /* Reject tracks that lack required tags */
164 if(*required_tags && !tag_intersection(track_tags, required_tags))
165 return 0;
166
36a1a882
RK
167 /* Use the configured weight if available */
168 if((s = kvp_get(prefs, "weight"))) {
169 long n;
170 errno = 0;
171
172 n = strtol(s, 0, 10);
173 if((errno == 0 || errno == ERANGE) && n >= 0)
174 return n;
175 }
05dcfac6
RK
176
177 /* Bias up tracks that were recently added */
178 if((s = kvp_get(data, "_noticed"))) {
179 const time_t noticed = atoll(s);
180
181 if(noticed + config->new_bias_age < now)
182 /* Currently we just step up the weight of tracks that are in range. A
183 * more sophisticated approach would be to linearly decay from new_bias
184 * down to BASE_WEIGHT over the course of the new_bias_age interval
185 * starting when the track is added. */
186 return config->new_bias;
187 }
36a1a882 188
05dcfac6 189 return BASE_WEIGHT;
bea6f6d5
RK
190}
191
fd9000bc
RK
192static unsigned char random_buffer[4096];
193static size_t random_left;
bea6f6d5 194
fd9000bc
RK
195/** @brief Fill [buf, buf+n) with random bytes */
196static void random_bytes(unsigned char *buf, size_t n) {
197 while(n > 0) {
198 if(random_left > 0) {
199 const size_t this_time = n > random_left ? random_left : n;
200
201 memcpy(buf, random_buffer + random_left - this_time, this_time);
202 n -= this_time;
203 random_left -= this_time;
204 } else {
205 static int fd = -1;
206 int r;
207
208 if(fd < 0) {
209 if((fd = open("/dev/urandom", O_RDONLY)) < 0)
210 fatal(errno, "opening /dev/urandom");
211 }
212 if((r = read(fd, random_buffer, sizeof random_buffer)) < 0)
213 fatal(errno, "reading /dev/urandom");
214 if((size_t)r < sizeof random_buffer)
215 fatal(0, "short read from /dev/urandom");
216 random_left = sizeof random_buffer;
217 }
bea6f6d5 218 }
4005a0d6 219}
220
221/** @brief Pick a random integer uniformly from [0, limit) */
222static unsigned long long pick_weight(unsigned long long limit) {
223 unsigned char buf[(sizeof(unsigned long long) * CHAR_BIT + 7)/8], m;
224 unsigned long long t, r, slop;
225 int i, nby, nbi;
226
227 //info("pick_weight: limit = %llu", limit);
228
229 /* First, decide how many bits of output we actually need; do bytes first
230 * (they're quicker) and then bits.
231 *
232 * To speed this up, we could use a binary search if we knew where to
233 * start. (Note that shifting by ULLONG_BITS or more (if such a constant
234 * existed) is undefined behaviour, so we mustn't do that.) Figuring out a
235 * start point involves preprocessor and/or autoconf magic.
236 */
237 for (nby = 1, t = (limit - 1) >> 8; t; nby++, t >>= 8)
238 ;
239 nbi = (nby - 1) << 3; t = limit >> nbi;
240 if (t >> 4) { t >>= 4; nbi += 4; }
241 if (t >> 2) { t >>= 2; nbi += 2; }
242 if (t >> 1) { t >>= 1; nbi += 1; }
243 nbi++;
244 //info("nby = %d; nbi = %d", nby, nbi);
245
246 /* Main randomness collection loop. We read a number of bytes from the
247 * randomness source, and glue them together into an integer (dropping
248 * bits off the top byte as necessary). Call the result r; we have
249 * 2^{nbi - 1) <= limit < 2^nbi and r < 2^nbi. If r < limit then we win;
250 * otherwise we try again. Given the above bounds, we expect fewer than 2
251 * iterations.
252 *
253 * Unfortunately there are subtleties. In particular, 2^nbi may in fact be
254 * zero due to overflow. So in fact what we do is compute slop = 2^nbi -
255 * limit > 0; if r < slop then we try again, otherwise r - slop is our
256 * winner.
257 */
258 slop = (2 << (nbi - 1)) - limit;
259 m = nbi & 7 ? (1 << (nbi & 7)) - 1 : 0xff;
260 //info("slop = %llu", slop);
261 //info("m = 0x%02x", m);
262
263 do {
264 /* Actually get some random data. */
265 random_bytes(buf, nby);
266
267 /* Clobber the top byte. */
268 buf[0] &= m;
269
270 /* Turn it into an integer. */
271 for (r = 0, i = 0; i < nby; i++)
272 r = (r << 8) | buf[i];
273 //info("r = %llu", r);
274 } while (r < slop);
275
276 return r - slop;
bea6f6d5
RK
277}
278
74b1f70d 279/** @brief Called for each track */
280static int collect_tracks_callback(const char *track,
281 struct kvp *data,
282 struct kvp *prefs,
283 void attribute((unused)) *u,
284 DB_TXN attribute((unused)) *tid) {
285 unsigned long weight = compute_weight(track, data, prefs);
bea6f6d5 286
74b1f70d 287 /* Decide whether this is the winning track.
288 *
289 * Suppose that we have n things, and thing i, for 0 <= i < n, has weight
290 * w_i. Let c_i = w_0 + ... + w_{i-1} be the cumulative weight of the
291 * things previous to thing i, and let W = c_n = w_0 + ... + w_{i-1} be the
292 * total weight. We can clearly choose a random thing with the correct
293 * weightings by picking a random number r in [0, W) and chooeing thing i
294 * where c_i <= r < c_i + w_i. But this involves having an enormous list
295 * and taking two passes over it (which has bad locality and is ugly).
296 *
297 * Here's another way. Initialize v = -1. Examine the things in order;
298 * for thing i, choose a random number r_i in [0, c_i + w_i). If r_i < w_i
299 * then set v <- i.
300 *
301 * Claim. For all 0 <= i < n, the above algorithm chooses thing i with
302 * probability w_i/W.
303 *
304 * Proof. Induction on n. The claim is clear for n = 1. Suppose it's
305 * true for n - 1. Let L be the event that we choose thing n - 1. Clearly
306 * Pr[L] = w_{n-1}/W. Condition on not-L: then the probabilty that we
307 * choose thing i, for 0 <= i < n - 1, is w_i/c_{n-1} (induction
308 * hypothesis); undoing the conditioning gives the desired result.
309 */
310 if(weight) {
311 total_weight += weight;
312 if (pick_weight(total_weight) < weight)
313 winning = track;
bea6f6d5 314 }
74b1f70d 315 ntracks++;
316 return 0;
bea6f6d5
RK
317}
318
319int main(int argc, char **argv) {
91c9324a
RK
320 int n, logsyslog = !isatty(2), err;
321 const char *tags;
bea6f6d5
RK
322
323 set_progname(argv);
324 mem_init();
325 if(!setlocale(LC_CTYPE, "")) fatal(errno, "error calling setlocale");
326 while((n = getopt_long(argc, argv, "hVc:dDSs", options, 0)) >= 0) {
327 switch(n) {
328 case 'h': help();
329 case 'V': version("disorder-choose");
330 case 'c': configfile = optarg; break;
331 case 'd': debugging = 1; break;
332 case 'D': debugging = 0; break;
333 case 'S': logsyslog = 0; break;
334 case 's': logsyslog = 1; break;
335 default: fatal(0, "invalid option");
336 }
337 }
338 if(logsyslog) {
339 openlog(progname, LOG_PID, LOG_DAEMON);
340 log_default = &log_syslog;
341 }
342 if(config_read(0)) fatal(0, "cannot read configuration");
28dc2d22
RK
343 /* Find out current queue/recent list */
344 queue_read();
345 recent_read();
bea6f6d5
RK
346 /* Generate the candidate track list */
347 trackdb_init(TRACKDB_NO_RECOVER);
348 trackdb_open(TRACKDB_NO_UPGRADE|TRACKDB_READ_ONLY);
349 global_tid = trackdb_begin_transaction();
91c9324a
RK
350 if((err = trackdb_get_global_tid("required-tags", global_tid, &tags)))
351 fatal(0, "error getting required-tags: %s", db_strerror(err));
352 required_tags = parsetags(tags);
353 if((err = trackdb_get_global_tid("prohibited-tags", global_tid, &tags)))
354 fatal(0, "error getting prohibited-tags: %s", db_strerror(err));
355 prohibited_tags = parsetags(tags);
bea6f6d5
RK
356 if(trackdb_scan(0, collect_tracks_callback, 0, global_tid))
357 exit(1);
358 trackdb_commit_transaction(global_tid);
359 trackdb_close();
360 trackdb_deinit();
361 //info("ntracks=%ld total_weight=%lld", ntracks, total_weight);
362 if(!total_weight)
363 fatal(0, "no tracks match random choice criteria");
74b1f70d 364 if(!winning)
365 fatal(0, "internal: failed to pick a track");
bea6f6d5 366 /* Pick a track */
74b1f70d 367 xprintf("%s", winning);
bea6f6d5
RK
368 xfclose(stdout);
369 return 0;
370}
371
372/*
373Local Variables:
374c-basic-offset:2
375comment-column:40
376fill-column:79
377indent-tabs-mode:nil
378End:
379*/