Roxen.git/server/base_server/cache.pike:28fdbe12ae24d05960370f8351b73a889a683ce0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 // This file is part of Roxen WebServer. // Copyright © 1996 - 2009, Roxen IS. // $Id: cache.pike,v 1.144 2011/04/15 11:50:31 mast Exp $ // FIXME: Add argcache, imagecache & protcache #include <roxen.h> #include <config.h> #ifdef MORE_CACHE_DEBUG # define MORE_CACHE_WERR(X...) report_debug("CACHE: "+X) # undef CACHE_DEBUG # define CACHE_DEBUG #else # define MORE_CACHE_WERR(X...) 0 #endif #ifdef CACHE_DEBUG # define CACHE_WERR(X...) report_debug("CACHE: "+X) #else # define CACHE_WERR(X...) 0 #endif int total_size_limit = 1024 * 1024; //! Maximum cache size for all cache managers combined. Start out with //! a bit of space before the setting is read. constant rebalance_adaptivity = 30 * 60; //! Number of seconds it should take to update the size balance //! between the cache managers for a different workload. //! //! Note: If this changes then the help texts in //! config_interface/actions/cachestatus.pike should be updated. constant rebalance_interval = 10; //! Seconds between @[update_cache_size_balance] calls. constant rebalance_keep_factor = 1.0 - 1.0 / (rebalance_adaptivity / rebalance_interval); //! Every time @[update_cache_size_balance] is called, the size //! assigned to each cache manager is allowed to shrink to no more //! than its currently assigned size multiplied by this factor. //! //! @note //! It's possible that cache managers are rebalanced a lot quicker //! than this: If a manager overshoots in one iteration then all //! caches are shrunk linearly in the next, and that is not limited by //! this value. Remains to be seen whether that is a feature or a bug. constant rebalance_min_size = 1024 * 1024; //! Minimum size to give a cache manager. protected constant cm_stats_avg_period = rebalance_adaptivity; // Approximate time period for which the decaying average/sum stats in // CacheManager apply. void set_total_size_limit (int size) //! Sets the total size limit available to all caches. { total_size_limit = size; // Rebalance immediately after setting the total size limit. This is // important mostly at server startup since modules otherwise might // do cache-intensive processing before the cache manager size has // been raised from its minimum value, resulting in sub-optimal // performance or even halting cache-filler operations // (e.g. Sitebuilder's workarea prefetcher.) update_cache_size_balance(); } class CacheEntry (mixed key, mixed data) //! Base class for cache entries. { // FIXME: Consider unifying this with CacheKey. But in that case we // need to ensure "interpreter lock" atomicity below. int size; //! The size of this cache entry, as measured by @[Pike.count_memory]. #ifdef DEBUG_CACHE_SIZES int cmp_size; // Size without counting strings. Used to compare the size between // cache_set and cache_clean. Strings are excluded since they might // get or lose unrelated refs in the time between which would make // the comparison unreliable. This might make us miss significant // strings though, but it's hard to get around it. #endif int timeout; //! Unix time when the entry times out, or zero if there's no //! timeout. //! @decl int|float cost; //! //! The creation cost for the entry, according to the metric used by //! the cache manager (provided it implements cost). protected string format_key() { if (stringp (key)) { if (sizeof (key) > 40) return sprintf ("%q...", key[..39 - sizeof ("...")]); else return sprintf ("%q", key); } else if (intp (key) || floatp (key) || objectp (key)) return sprintf ("%O", key); else return sprintf ("%t", key); } protected string _sprintf (int flag) { return flag == 'O' && sprintf ("CacheEntry(%s, %db)", format_key(), size); } } class CacheStats //! Holds statistics for each named cache. { int count; //! The number of entries in the cache. int size; //! The sum of @[CacheEntry.size] for all cache entries in the cache. int hits, misses; //! Plain counts of cache hits and misses since the creation of the //! cache. int|float cost_hits, cost_misses; //! Hit and miss count according to the cache manager cost metric //! since the creation of the cache. Note that @[cost_misses] is //! determined when a new entry is added - it will not include when //! no new entry was created after a cache miss. #ifdef CACHE_BYTE_HR_STATS int byte_hits, byte_misses; //! Byte hit and miss count since the creation of the cache. Note //! that @[byte_misses] is determined when a new entry is added - it //! will not include when no new entry was created after a cache //! miss. #endif protected string _sprintf (int flag) { return flag == 'O' && sprintf ("CacheStats(%d, %dk)", count, size / 1024); } } class CacheManager //! A cache manager handles one or more caches, applying the same //! eviction policy and the same size limit on all of them. I.e. it's //! practically one cache, and the named caches inside only act as //! separate name spaces. { //! @decl constant string name; //! //! A unique name to identify the manager. It is also used as //! display name. //! @decl constant string doc; //! //! A description of the manager and its eviction policy in html. constant has_cost = 0; //! Nonzero if this cache manager implements a cost metric. int total_size_limit = global::total_size_limit; //! Maximum allowed size including the cache manager overhead. int size; //! The sum of @[CacheStats.size] for all named caches. int size_limit = global::total_size_limit; //! Maximum allowed size for cache entries - @[size] should never //! greater than this. This is a cached value calculated from //! @[total_size_limit] on regular intervals. int entry_add_count; //! Number of entries added since this cache manager was created. mapping(string:mapping(mixed:CacheEntry)) lookup = ([]); //! Lookup mapping on the form @expr{(["cache_name": ([key: data])])@}. //! //! For functions in this class, a cache submapping does not exist //! only due to race, so the cache should just be ignored. mapping(string:CacheStats) stats = ([]); //! Statistics for the named caches managed by this object. //! //! For functions in this class, a @[CacheStats] object does not //! exist only due to race, so the cache should just be ignored. //! @decl program CacheEntry; //! //! The manager-specific class to use to create @[CacheEntry] objects. void clear_cache_context() {} //! Called to clear any thread-local state that the manager keeps to //! track execution times etc. This is called before a thread starts //! with a new request or other kind of job. void got_miss (string cache_name, mixed key, mapping cache_context); //! Called when @[cache_lookup] records a cache miss. protected void account_miss (string cache_name) { recent_misses++; if (CacheStats cs = stats[cache_name]) cs->misses++; } void got_hit (string cache_name, CacheEntry entry, mapping cache_context); //! Called when @[cache_lookup] records a cache hit. protected void account_hit (string cache_name, CacheEntry entry) { recent_hits++; recent_cost_hits += entry->cost; if (CacheStats cs = stats[cache_name]) { cs->hits++; cs->cost_hits += entry->cost; #ifdef CACHE_BYTE_HR_STATS cs->byte_hits += entry->size; #endif } } int add_entry (string cache_name, CacheEntry entry, int old_entry, mapping cache_context); //! Called to add an entry to the cache. Should also evict entries //! as necessary to keep @expr{@[size] <= @[size_limit]@}. //! //! If @[old_entry] is set then the entry hasn't been created from //! scratch, e.g. there is no prior @[got_miss] call. Returns 1 if //! the entry got added to the cache. Returns 0 if the function //! chose to evict it immediately or if the cache has disappeared. private void account_remove_entry (string cache_name, CacheEntry entry) { if (CacheStats cs = stats[cache_name]) { cs->count--; cs->size -= entry->size; ASSERT_IF_DEBUG (cs->size /*%O*/ >= 0, cs->size); ASSERT_IF_DEBUG (cs->count /*%O*/ >= 0, cs->count); } size -= entry->size; ASSERT_IF_DEBUG (size /*%O*/ >= 0, size); } protected int low_add_entry (string cache_name, CacheEntry entry) { ASSERT_IF_DEBUG (entry->size /*%O*/, entry->size); // Assume that the addition of the new entry came about due to a // cache miss. recent_cost_misses += entry->cost; if (CacheStats cs = stats[cache_name]) { cs->cost_misses += entry->cost; #ifdef CACHE_BYTE_HR_STATS cs->byte_misses += entry->size; #endif if (mapping(mixed:CacheEntry) lm = lookup[cache_name]) { // vvv Relying on the interpreter lock from here. CacheEntry old_entry = lm[entry->key]; lm[entry->key] = entry; // ^^^ Relying on the interpreter lock to here. if (old_entry) { account_remove_entry (cache_name, old_entry); recent_added_bytes -= entry->size; remove_entry (cache_name, old_entry); } cs->count++; cs->size += entry->size; size += entry->size; recent_added_bytes += entry->size; if (!(++entry_add_count & 0x3fff)) // = 16383 update_size_limit(); } return 1; } return 0; } int remove_entry (string cache_name, CacheEntry entry); //! Called to delete an entry from the cache. Should use //! @[low_remove_entry] to do the atomic removal. Must ensure the //! entry is removed from any extra data structures, regardless //! whether it's already gone from the @[lookup] mapping or not. //! Returns the return value from @[low_remove_entry]. protected int low_remove_entry (string cache_name, CacheEntry entry) //! Returns 1 if the entry got removed from the cache, or 0 if it //! wasn't found in the cache or if the cache has disappeared. { if (mapping(mixed:CacheEntry) lm = lookup[cache_name]) if (lm[entry->key] == entry) { // Relying on the interpreter here. m_delete (lm, entry->key); account_remove_entry (cache_name, entry); return 1; } return 0; } void evict (int max_size); //! Called to evict entries until @expr{@[size] <= @[max_size]@}. void after_gc() {} //! Called from the periodic GC, after stale and invalid entries //! have been removed from the cache. int manager_size_overhead() //! Returns the size consumed by the manager itself, excluding the //! cache entries. { int res = (Pike.count_memory (-1, this, lookup) + Pike.count_memory (0, stats)); foreach (lookup;; mapping(mixed:CacheEntry) lm) res += Pike.count_memory (-1, lm); return res; } float add_rate = 0.0; //! The number of newly added bytes per second, calculated as a //! decaying average over the last @[cm_stats_avg_period] seconds. //! Note that the returned value could become negative if entries //! are replaced with smaller ones. float hits = 0.0, misses = 0.0; //! Decaying sums over the hits and misses during approximately the //! last @[cm_stats_avg_period] seconds. float cost_hits = 0.0, cost_misses = 0.0; //! Decaying sums over the cost weighted hits and misses during //! approximately the last @[cm_stats_avg_period] seconds. Only //! applicable if @[has_cost] is set. protected int recent_added_bytes; protected int recent_hits, recent_misses; protected int|float recent_cost_hits, recent_cost_misses; void update_decaying_stats (int start_time, int last_update, int now) // Should only be called at regular intervals from // update_cache_size_balance. { // Skip updating if we did it recently (avoid division by zero below.) if (now == last_update) return; float last_period = (float) (now - last_update); float tot_period = (float) (now - start_time); int startup = tot_period < cm_stats_avg_period; if (!startup) tot_period = (float) cm_stats_avg_period; float our_weight = min (last_period / tot_period, 1.0); float old_weight = 1.0 - our_weight; if (startup) { hits += (float) recent_hits; misses += (float) recent_misses; cost_hits += (float) recent_cost_hits; cost_misses += (float) recent_cost_misses; } else { hits = old_weight * hits + (float) recent_hits; misses = old_weight * misses + (float) recent_misses; cost_hits = old_weight * cost_hits + (float) recent_cost_hits; cost_misses = old_weight * cost_misses + (float) recent_cost_misses; } add_rate = (old_weight * add_rate + our_weight * (recent_added_bytes / last_period)); recent_added_bytes = 0; recent_hits = recent_misses = 0; recent_cost_hits = recent_cost_misses = 0; } void update_size_limit() { MORE_CACHE_WERR ("%O: update_size_limit\n", this); int mgr_oh = manager_size_overhead(); size_limit = max (0, total_size_limit - mgr_oh); if (size > size_limit) { CACHE_WERR ("%O: Evicting %db " "(entry size limit %db, manager overhead %db, total %db)\n", this, size - size_limit, size_limit, mgr_oh, total_size_limit); evict (size_limit); } } int free_space() //! Returns the amount of unused space left. Might be negative in //! some narrow time windows when the cache is over its limit and //! @[evict] hasn't yet catched up. { return size_limit - size; } string format_cost (int|float cost) {return "-";} //! Function to format a cost measurement for display in the status //! page. protected string _sprintf (int flag) { return flag == 'O' && sprintf ("CacheManager(%s: %dk/%dk)", this->name || "-", size / 1024, size_limit / 1024); } } #if 0 class CM_Random { inherit CacheManager; constant name = "Random"; constant doc = #"\ This is a very simple cache manager that just evicts entries from the cache at random. The only upside with it is that the cache management overhead is minimal."; // Workaround since "constant CacheEntry = global::CacheEntry;" // currently causes segfault in 7.8. constant CacheEntry = global::CacheEntry; void got_miss (string cache_name, mixed key, mapping cache_context) { account_miss (cache_name); } void got_hit (string cache_name, CacheEntry entry, mapping cache_context) { account_hit (cache_name, entry); } int add_entry (string cache_name, CacheEntry entry, int old_entry, mapping cache_context) { int res = low_add_entry (cache_name, entry); if (size > size_limit) evict (size_limit); return res; } int remove_entry (string cache_name, CacheEntry entry) { return low_remove_entry (cache_name, entry); } void evict (int max_size) { while (size > max_size) { if (!sizeof (lookup)) break; // Relying on the interpreter lock here. string cache_name = random (lookup)[0]; if (mapping(mixed:CacheEntry) lm = lookup[cache_name]) { if (sizeof (lm)) { // Relying on the interpreter lock here. CacheEntry entry = random (lm)[1]; MORE_CACHE_WERR ("evict: Size %db > %db - evicting %O / %O.\n", size, max_size, cache_name, entry); low_remove_entry (cache_name, entry); } else m_delete (lookup, cache_name); } } } } protected CM_Random cm_random = CM_Random(); #endif class CM_GreedyDual //! Base class for cache managers that works with some variant of the //! GreedyDual algorithm (see e.g. Cao and Irani, "Cost-Aware WWW //! Proxy Caching Algorithms" in "Proceedings of the 1997 USENIX //! Symposium on Internet Technology and Systems"): //! //! A priority queue is maintained, which contains all entries ordered //! according to a priority value. The entry with the lowest priority //! is always chosen for eviction. When a new entry p is added, and //! each time it is hit afterwards, the priority is set to v(p) + L, //! where v(p) is p's value according to some algorithm-specific //! definition, and L is the lowest priority value in the cache. This //! means that the priority values constantly increases, so that old //! entries without hits eventually gets evicted regardless of their //! initial v(p). { inherit CacheManager; class CacheEntry { inherit global::CacheEntry; int|float value; //! The value of the entry, i.e. v(p) in the class description. int|float pval; //! The priority value for the entry, defining its position in //! @[priority_list]. Must not change for an entry that is //! currently a member of @[priority_list]. string cache_name; //! Need the cache name to find the entry, since @[priority_list] //! is global. protected int `< (CacheEntry other) { return pval < other->pval; } protected string _sprintf (int flag) { return flag == 'O' && sprintf ("CM_GreedyDual.CacheEntry(%O: %s, %db, %O)", pval, format_key(), size, value); } } multiset(CacheEntry) priority_list = (<>); //! A list of all entries in priority order, by using the multiset //! builtin sorting through @[CacheEntry.`<]. protected int max_used_pval; // Used to detect when the entry pval's get too big to warrant a // reset. // // For integers, this is the maximum used pval and we reset at the // next gc when it gets over Int.NATIVE_MAX/2 (to have some spare // room for the gc delay). // // For floats, we reset whenever L is so big that less than 8 // significant bits remains when v(p) is added to it. In that case // max_used_pval only works as a flag, and we set it to // Int.NATIVE_MAX when that state is reached. #ifdef CACHE_DEBUG protected void debug_check_priority_list() // Assumes no concurrent access - run inside _disable_threads. { werror ("Checking priority_list with %d entries.\n", sizeof (priority_list)); CacheEntry prev; foreach (priority_list; CacheEntry entry;) { if (!prev) prev = entry; else if (prev >= entry) error ("Found cache entries in wrong order: %O vs %O in %O\n", prev, entry, priority_list); if (intp (entry->pval) && entry->pval > max_used_pval) error ("Found %O with pval higher than max %O.\n", entry, max_used_pval); } } #endif int|float calc_value (string cache_name, CacheEntry entry, int old_entry, mapping cache_context); //! Called to calculate the value for @[entry], which gets assigned //! to the @expr{value@} variable. Arguments are the same as to //! @[add_entry]. void got_miss (string cache_name, mixed key, mapping cache_context) { account_miss (cache_name); } local protected int|float calc_pval (CacheEntry entry) { int|float pval; if (CacheEntry lowest = get_iterator (priority_list)->index()) { int|float l = lowest->pval, v = entry->value; pval = l + v; if (floatp (v)) { if (v != 0.0 && v < l * (Float.EPSILON * 0x10)) { #ifdef DEBUG if (max_used_pval != Int.NATIVE_MAX) werror ("%O: Ran out of significant digits for cache entry %O - " "got min priority %O and entry value %O.\n", this, entry, l, v); #endif // Force a reset of the pvals in the next gc. max_used_pval = Int.NATIVE_MAX; } } else if (pval > max_used_pval) max_used_pval = pval; } else // Assume entry->value isn't greater than Int.NATIVE_MAX/2 right away. pval = entry->value; return pval; } void got_hit (string cache_name, CacheEntry entry, mapping cache_context) { account_hit (cache_name, entry); int|float pval = calc_pval (entry); // Note: Won't have the interpreter lock during the operation below. The // tricky situation occur if the same entry is processed by another // got_hit, where we might get it inserted in more than one place and one // of them will be inconsistent with the pval value. evict() has a // consistency check that should detect and correct this eventually. The // worst thing that happens is that the eviction order is more or less // wrong until then. priority_list[entry] = 0; entry->pval = pval; priority_list[entry] = 1; } int add_entry (string cache_name, CacheEntry entry, int old_entry, mapping cache_context) { entry->cache_name = cache_name; int|float v = entry->value = calc_value (cache_name, entry, old_entry, cache_context); if (!low_add_entry (cache_name, entry)) return 0; entry->pval = calc_pval (entry); priority_list[entry] = 1; if (size > size_limit) evict (size_limit); return 1; } int remove_entry (string cache_name, CacheEntry entry) { priority_list[entry] = 0; return low_remove_entry (cache_name, entry); } void evict (int max_size) { object threads_disabled; while (size > max_size) { CacheEntry entry = get_iterator (priority_list)->index(); if (!entry) break; if (!priority_list[entry]) { // The order in the list has become inconsistent with the pval values. // It might happen due to the race in got_hit(), or it might just be // interference from a concurrent evict() call. if (!threads_disabled) // Take a hefty lock and retry, to rule out the second case. threads_disabled = _disable_threads(); else { // Got the lock so it can't be a race, i.e. the priority_list order // is funky. Have to rebuild it without interventions. #ifdef CACHE_DEBUG debug_check_priority_list(); #endif report_warning ("Warning: Recovering from race inconsistency " "in %O->priority_list (on %O).\n", this, entry); priority_list = (<>); foreach (lookup; string cache_name; mapping(mixed:CacheEntry) lm) foreach (lm;; CacheEntry entry) priority_list[entry] = 1; } continue; } priority_list[entry] = 0; MORE_CACHE_WERR ("evict: Size %db > %db - evicting %O / %O.\n", size, max_size, entry->cache_name, entry); low_remove_entry (entry->cache_name, entry); } threads_disabled = 0; } int manager_size_overhead() { return Pike.count_memory (-1, priority_list) + ::manager_size_overhead(); } void after_gc() { if (max_used_pval > Int.NATIVE_MAX / 2) { int|float pval_base; if (CacheEntry lowest = get_iterator (priority_list)->index()) pval_base = lowest->pval; else return; // To cope with concurrent updates, we replace the lookup // mapping and start adding back the entries from the old one. // Need _disable_threads to make the resets of the CacheStats // fields atomic. object threads_disabled = _disable_threads(); mapping(string:mapping(mixed:CacheEntry)) old_lookup = lookup; lookup = ([]); foreach (old_lookup; string cache_name;) { lookup[cache_name] = ([]); if (CacheStats cs = stats[cache_name]) { cs->count = 0; cs->size = 0; } } priority_list = (<>); size = 0; max_used_pval = 0; threads_disabled = 0; int|float max_pval; foreach (old_lookup; string cache_name; mapping(mixed:CacheEntry) old_lm) if (CacheStats cs = stats[cache_name]) if (mapping(mixed:CacheEntry) new_lm = lookup[cache_name]) foreach (old_lm; mixed key; CacheEntry entry) { int|float pval = entry->pval -= pval_base; if (!new_lm[key]) { // Relying on the interpreter lock here. new_lm[key] = entry; priority_list[entry] = 1; if (pval > max_pval) max_pval = pval; cs->count++; cs->size += entry->size; size += entry->size; } } #ifdef CACHE_DEBUG debug_check_priority_list(); #endif if (intp (max_pval)) max_used_pval = max_pval; CACHE_WERR ("%O: Rebased priority values - " "old base was %O, new range is 0..%O.\n", this, pval_base, max_pval); if (Configuration admin_config = roxenp()->get_admin_configuration()) // Log an event, in case we suspect this starts to happen a lot. admin_config->log_event ("roxen", "ram-cache-rebase", this->name, ([ "old-pval-base": pval_base, "new-max-pval": max_pval, ])); } } } class CM_GDS_1 { inherit CM_GreedyDual; constant name = "GDS(1)"; constant doc = #"\ This cache manager implements <a href='http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.30.7285'>GreedyDual-Size</a> with the cost of each entry fixed at 1, which makes it optimize the cache hit ratio."; float calc_value (string cache_name, CacheEntry entry, int old_entry, mapping cache_context) { ASSERT_IF_DEBUG (entry->size /*%O*/ > 10, entry->size); return 1.0 / entry->size; } } protected CM_GDS_1 cm_gds_1 = CM_GDS_1(); class CM_GDS_Time //! Like @[CM_GDS_1] but adds support for calculating entry cost based //! on passed time. { inherit CM_GreedyDual; constant has_cost = 1; class CacheEntry { inherit CM_GreedyDual::CacheEntry; int|float cost; protected string _sprintf (int flag) { return flag == 'O' && sprintf ("CM_GDS_Time.CacheEntry(%O: %s, %db, %O, %O)", pval, format_key(), size, value, cost); } } protected Thread.Local cache_contexts = Thread.Local(); // A thread local mapping to store the timestamp from got_miss so it // can be read from the (presumably) following add_entry. // // In an entry with index 0 in the mapping, the time spent creating // cache entries is accumulated. It is used to deduct the time for // creating entries in subcaches. void clear_cache_context() { cache_contexts->set (([])); } protected int gettime_func(); //! Returns the current time for cost calculation. (@[format_cost] //! assumes this is in microseconds.) protected void save_start_hrtime (string cache_name, mixed key, mapping cache_context) { if (mapping all_ctx = cache_context || cache_contexts->get()) { int start = gettime_func() - all_ctx[0]; if (mapping(mixed:int) ctx = all_ctx[cache_name]) { #if 0 if (!zero_type (ctx[key])) // This warning is useful since strictly speaking we don't // know which cache_lookup calls to use as start for the // time measurement, so the time cost might be bogus. If it // isn't the last one then you should probably replace some // calls with cache_peek. werror ("Warning: Detected repeated missed lookup calls.\n%s\n", describe_backtrace (backtrace())); #endif ctx[key] = start; } else all_ctx[cache_name] = ([key: start]); } else { #ifdef DEBUG werror ("Warning: Got call from %O without cache context mapping.\n%s\n", Thread.this_thread(), describe_backtrace (backtrace())); #endif } } void got_miss (string cache_name, mixed key, mapping cache_context) { //werror ("Miss.\n%s\n", describe_backtrace (backtrace())); account_miss (cache_name); save_start_hrtime (cache_name, key, cache_context); } void got_hit (string cache_name, CacheEntry entry, mapping cache_context) { account_hit (cache_name, entry); // It shouldn't be necessary to record the start time for cache // hits, but do it anyway for now since there are caches that on // cache hits extend the entries with more data. save_start_hrtime (cache_name, entry->key, cache_context); } protected int entry_create_hrtime (string cache_name, mixed key, mapping cache_context) { if (mapping all_ctx = cache_context || cache_contexts->get()) if (mapping(mixed:int) ctx = all_ctx[cache_name]) { int start = m_delete (ctx, key); if (!zero_type (start)) { int duration = (gettime_func() - all_ctx[0]) - start; #if 0 // This assertion is disabled for now, since it doesn't work // correctly when entry creations are interleaved instead of // properly nested. // // However, handling that would mean more overhead. Just to // detect interleaving we need to store the real timestamp // in save_start_hrtime, in addition to the one with // all_ctx[0] subtracted. And to evenly distribute the // overlapping time between the interleaved entries, we'd // have to track them and update their creation costs // afterwards. If all that work is concentrated to this // function (to keep save_start_hrtime as lean as possible // since it's called much more often), it'd be an O(n^2) // process. // // Also note that this assertion might trig if gettime_func // isn't monotonic (c.f. FIXME in CM_GDS_RealTime.gettime_func). ASSERT_IF_DEBUG (duration /*%O*/ >= 0 /* start: %O, all_ctx: %O */, duration, start, all_ctx); #endif if (duration < 0) // Can get negative duration when entry creation is // interleaved. Consider this case (t_acc is all_ctx[0]): // // 0 t5 // Create entry A: |----------------------------| // t1 t3 // Create entry B: |----------------| // t_acc == 0 ^ t2 t4 // Create entry C: |----------| // t_acc == 0 ^ ^ t_acc == t3 // // When we get here for entry C, the accumulated time is // t3, which could be larger than t4 - t2, thereby causing // negative duration. Setting the duration to 0 here not // only affects this entry; it also gives too much time // (t4 - t3) to the entry A which encompasses both // interleaved entries. The error stops there, though. duration = 0; all_ctx[0] += duration; return duration; } } #ifdef DEBUG // Note that this also happens if the caller calls cache_set() // several times to add the same entry. That should be avoided // (see the docs for cache_set). werror ("Warning: No preceding lookup for this key - " "cannot determine entry creation time.\n%s\n", describe_backtrace (backtrace())); #endif return 0; } protected mapping(string:array(float|int)) mean_costs = ([]); // Stores the mean cost for all entries in each cache. The values // are on the form ({mean, count}), where count is the number of // samples that have contributed to the mean. These are not real // mean values since we (normally) don't keep track of the cost of // each entry. Instead it's a decaying average, where count is // capped at 1000. // // FIXME: Nowadays we actually do keep track of the cost for each // entry. float calc_value (string cache_name, CacheEntry entry, int old_entry, mapping cache_context) { if (int hrtime = !old_entry && entry_create_hrtime (cache_name, entry->key, cache_context)) { float cost = entry->cost = (float) hrtime; if (array(float|int) mean_entry = mean_costs[cache_name]) { [float mean_cost, int mean_count] = mean_entry; mean_entry[0] = (mean_count * mean_cost + cost) / (mean_count + 1); if (mean_count < 1000) mean_entry[1] = mean_count + 1; } else mean_costs[cache_name] = ({cost, 1}); return cost / entry->size; } // Awkward situation: We don't have any cost for this entry. Just // use the mean cost of all entries in the cache, so it at least // isn't way off in either direction. if (array(float|int) mean_entry = mean_costs[cache_name]) return mean_entry[0] / entry->size; else return 0.0; // Here goes nothing.. :P } void evict (int max_size) { ::evict (max_size); if (!max_size) mean_costs = ([]); } string format_cost (float cost) { return Roxen.format_hrtime ((int) cost); } } class CM_GDS_CPUTime { inherit CM_GDS_Time; constant name = "GDS(cpu time)"; string doc = #"\ This cache manager implements <a href='http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.30.7285'>GreedyDual-Size</a> with the cost of each entry determined by the CPU time it took to create it. The CPU time implementation is " + Roxen.html_encode_string (System.CPU_TIME_IMPLEMENTATION) + " which is " + (System.CPU_TIME_IS_THREAD_LOCAL ? "thread local" : "not thread local") + " and has a resolution of " + (System.CPU_TIME_RESOLUTION / 1e6) + " ms."; protected int gettime_func() { return gethrvtime(); } } protected CM_GDS_CPUTime cm_gds_cputime = CM_GDS_CPUTime(); class CM_GDS_RealTime { inherit CM_GDS_Time; constant name = "GDS(real time)"; string doc = #"\ This cache manager implements <a href='http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.30.7285'>GreedyDual-Size</a> with the cost of each entry determined by the real (wall) time it took to create it. The real time implementation is " + Roxen.html_encode_string (System.REAL_TIME_IMPLEMENTATION) + " which is " + (System.REAL_TIME_IS_MONOTONIC ? "monotonic" : "not monotonic") + " and has a resolution of " + (System.REAL_TIME_RESOLUTION / 1e6) + " ms."; protected int gettime_func() { // The real time includes a lot of noise that isn't appropriate // for cache entry cost measurement. Let's compensate for the time // spent in the pike gc, at least. // // FIXME: This function isn't monotonic if it's used inside gc // runs. That should be rare though, so we ignore it for now. return gethrtime() - Pike.implicit_gc_real_time(); } } protected CM_GDS_RealTime cm_gds_realtime = CM_GDS_RealTime(); //! The preferred managers according to various caching requirements. //! When several apply for a cache, choose the first one in this list. //! //! @dl //! @item "default" //! The default manager for caches that do not specify any //! requirements. //! //! @item "no_cpu_timings" //! The manager to use for caches where a cache entry is created //! synchronously by one thread, but that thread spends most of its //! time waiting for a result from an external party, meaning that //! consumed cpu time is not an accurate measurement of the cost. //! //! @item "no_thread_timings" //! The manager to use for caches where a cache entry isn't created //! synchronously by one thread in the span between the //! @[cache_lookup] miss and the following @[cache_set]. //! //! @item "no_timings" //! The manager to use for caches that do not have a usage pattern //! where it is meaningful to calculate the creation cost from the //! time between a @[cache_lookup] miss to the following //! @[cache_set]. //! @enddl mapping(string:CacheManager) cache_manager_prefs = ([ #if 0 // Workaround: The following uses GDS(cpu time) for most caches, // which should be more accurate. But since several caches cannot // use it, we get a balancing problem between it and GDS(real time). // Because update_cache_size_balance still isn't good enough at // responding to workload changes quickly, we stick to GDS(real // time) for now to make this problem less significant (there might // also be a few caches in GDS(1), but it's not large enough to // really matter). "default": (System.CPU_TIME_IS_THREAD_LOCAL != "yes" || System.CPU_TIME_RESOLUTION > 10000 ? // Don't use cpu time if it's too bad. Buglet: We just // assume the real time is better. cm_gds_realtime : cm_gds_cputime), #else "default": cm_gds_realtime, #endif "no_cpu_timings": cm_gds_realtime, "no_thread_timings": cm_gds_realtime, "no_timings": cm_gds_1, ]); //! All available cache managers. array(CacheManager) cache_managers = Array.uniq (({cache_manager_prefs->default, cache_manager_prefs->no_cpu_timings, cache_manager_prefs->no_thread_timings, cache_manager_prefs->no_timings, })); protected Thread.Mutex cache_mgmt_mutex = Thread.Mutex(); // Locks operations that manipulate named caches, i.e. changes in the // caches, CacheManager.stats and CacheManager.lookup mappings. protected int cache_start_time = time() - 1; protected int last_cache_size_balance = time() - 1; // Subtract 1 from the initial values to avoid division by zero in // update_decaying_stats if update_cache_size_balance gets called // really soon after startup. protected void update_cache_size_balance() //! Updates the balance between the fractions of the total size //! allocated to each cache manager. { Thread.MutexKey lock = cache_mgmt_mutex->lock(); int now = time(); mapping(CacheManager:int) mgr_used = ([]); int used_size; // Update the decaying sums in the CacheStats objects. foreach (cache_managers, CacheManager mgr) mgr->update_decaying_stats (cache_start_time, last_cache_size_balance, now); if (!total_size_limit) { // Caches are effectively disabled. Ignore rebalance_min_size in this case. foreach (cache_managers, CacheManager mgr) { mgr->total_size_limit = 0; mgr->update_size_limit(); } } else { foreach (cache_managers, CacheManager mgr) used_size += mgr_used[mgr] = mgr->size + (mgr->total_size_limit - mgr->size_limit); if (used_size < (int) (0.9 * total_size_limit)) { // The caches are underpopulated. Set the limits so that they all can // grow freely until the total limit is reached. The way this is done // might cause the combined size overshoot the limit, but we assume // update_cache_size_balance runs often enough to not let it get // seriously out of hand. int extra = total_size_limit - used_size; CACHE_WERR ("Rebalance: %db under the limit - free growth " "(overshoot %db).\n", extra, extra * sizeof (cache_managers) + used_size - total_size_limit); foreach (cache_managers, CacheManager mgr) { mgr->total_size_limit = max (rebalance_min_size, mgr_used[mgr] + extra); mgr->update_size_limit(); } } else { int reserved_size; #ifdef CACHE_DEBUG int overshoot_size; #endif mapping(CacheManager:int) mgr_reserved_size = ([]); mapping(CacheManager:float) mgr_weights = ([]); float total_weight = 0.0; foreach (cache_managers, CacheManager mgr) { int reserved = (int) (rebalance_keep_factor * mgr_used[mgr]); if (reserved < rebalance_min_size) { // Don't reserve more than the used size - the rest is up for grabs // by whichever cache fills it first. If that's under // rebalance_min_size it means we get some "overshoot" room here // too, like above. reserved = mgr_used[mgr]; #ifdef CACHE_DEBUG overshoot_size += rebalance_min_size - reserved; #endif } reserved_size += mgr_reserved_size[mgr] = reserved; float lookups = mgr->has_cost ? mgr->cost_hits + mgr->cost_misses : mgr->hits + mgr->misses; float hit_rate_per_byte = lookups != 0.0 && mgr_used[mgr] ? mgr->hits / lookups / mgr_used[mgr] : 0.0; // add_rate is a measurement on how many new bytes a cache could put // into use, and hit_rate_per_byte weighs in a projection on how // successful it would be caching those bytes. float weight = max (mgr->add_rate * hit_rate_per_byte, 0.0); mgr_weights[mgr] = weight; total_weight += weight; CACHE_WERR ("Rebalance weight %s: Reserved %db, " "add rate %g, hr %g, hr/b %g, weight %g.\n", mgr->name, reserved, mgr->add_rate, lookups != 0.0 ? mgr->hits / lookups : 0.0, hit_rate_per_byte, weight); } if (total_weight > 0.0) { // Don't change anything if there's no weight on any cache // manager. That means there's no activity. if (reserved_size > total_size_limit) { // The caches are over the limit so much there's no room for // weighted redistribution. Either total_size_limit has shrunk or // they have overshot it as a result of the free growth policy. // Shrink each cache manager relative to its current size. // Ensure no manager gets below rebalance_min_size. Note that if a // cache uses less than rebalance_min_size then the remaining space // is available to the other caches, which means that sum of the // limits might get larger than total_size_limit. That's intentional // - it's a variant of the "overshoot" policy above. int total_above_min = total_size_limit; foreach (cache_managers, CacheManager mgr) { int reserved = min (mgr_used[mgr], rebalance_min_size); used_size -= reserved; mgr_used[mgr] -= reserved; total_above_min -= reserved; } if (used_size > 0) { CACHE_WERR ("Rebalance: %db over the limit - shrinking linearly " "(overshoot %db).\n", used_size - total_above_min, rebalance_min_size * sizeof (cache_managers) + total_above_min - total_size_limit); foreach (cache_managers, CacheManager mgr) { int new_size = rebalance_min_size + (int) (((float) mgr_used[mgr] / used_size) * total_above_min); CACHE_WERR ("Rebalance shrink %s: From %db to %db - diff %db.\n", mgr->name, mgr->total_size_limit, new_size, new_size - mgr->total_size_limit); mgr->total_size_limit = new_size; mgr->update_size_limit(); } } } else { int size_for_rebalance = total_size_limit - reserved_size; CACHE_WERR ("Rebalance: %db for rebalance (reserved %db, " "overshoot %db)\n", size_for_rebalance, reserved_size, overshoot_size); foreach (cache_managers, CacheManager mgr) { int new_size = max (mgr_reserved_size[mgr] + (int) ((mgr_weights[mgr] / total_weight) * size_for_rebalance), rebalance_min_size); CACHE_WERR ("Rebalance on weight %s: From %db to %db - diff %db.\n", mgr->name, mgr->total_size_limit, new_size, new_size - mgr->total_size_limit); mgr->total_size_limit = new_size; mgr->update_size_limit(); } } } } } #ifdef DEBUG foreach (cache_managers, CacheManager mgr) ASSERT_IF_DEBUG (mgr/*%O*/->total_size_limit /*%O*/ == 0 || mgr->total_size_limit >= rebalance_min_size, mgr, mgr->total_size_limit); #endif last_cache_size_balance = now; } protected void periodic_update_cache_size_balance() { update_cache_size_balance(); roxenp()->background_run (rebalance_interval, periodic_update_cache_size_balance); } #ifdef DEBUG_CACHE_SIZES protected int cmp_sizeof_cache_entry (string cache_name, CacheEntry entry) { int res; mixed data = entry->data; mapping opts = (["block_strings": 1, #if DEBUG_CACHE_SIZES > 1 "collect_internals": 1, #endif ]); if (function(int|mapping:int) cm_cb = objectp (data) && data->cache_count_memory) res = cm_cb (opts) + Pike.count_memory (-1, entry, entry->key); else res = Pike.count_memory (opts, entry, entry->key, data); #if DEBUG_CACHE_SIZES > 1 werror ("Internals counted for %O / %O: ({\n%{%s,\n%}})\n", cache_name, entry, sort (map (opts->collect_internals, lambda (mixed m) {return sprintf ("%O", m);}))); #endif return res; } #endif protected mapping(string:CacheManager) caches = ([]); // Maps the named caches to the cache managers that handle them. mapping(string:CacheManager) cache_list() //! Returns a list of all currently registered caches and their //! managers. { return caches + ([]); } CacheManager cache_register (string cache_name, void|string|CacheManager manager) //! Registers a new cache. Returns its @[CacheManager] instance. //! //! @[manager] can be a specific @[CacheManager] instance to use, a //! string that specifies a type of manager (see //! @[cache_manager_prefs]), or zero to select the default manager. //! //! If the cache already exists, its current manager is simply //! returned, and @[manager] has no effect. //! //! Registering a cache is not mandatory before it is used - one will //! be created automatically with the default manager otherwise. //! Still, it's a good idea so that the cache list in the admin //! interface gets populated timely. { Thread.MutexKey lock = cache_mgmt_mutex->lock (2); // Called from cache_change_manager too. if (CacheManager mgr = caches[cache_name]) return mgr; if (!manager) manager = cache_manager_prefs->default; else if (stringp (manager)) { string cache_type = manager; manager = cache_manager_prefs[cache_type]; if (!manager) error ("Unknown cache manager type %O requested.\n", cache_type); } caches[cache_name] = manager; manager->stats[cache_name] = CacheStats(); manager->lookup[cache_name] = ([]); return manager; } void cache_unregister (string cache_name) //! Unregisters the specified cache. This empties the cache and also //! removes it from the cache overview in the admin interface. { Thread.MutexKey lock = cache_mgmt_mutex->lock(); // vvv Relying on the interpreter lock from here. if (CacheManager mgr = m_delete (caches, cache_name)) { mapping(mixed:CacheEntry) lm = m_delete (mgr->lookup, cache_name); CacheStats cs = m_delete (mgr->stats, cache_name); // ^^^ Relying on the interpreter lock to here. mgr->size -= cs->size; destruct (lock); foreach (lm;; CacheEntry entry) mgr->remove_entry (cache_name, entry); } } CacheManager cache_get_manager (string cache_name) //! Returns the cache manager for the given cache, or zero if the //! cache isn't registered. { return caches[cache_name]; } void cache_change_manager (string cache_name, CacheManager manager) //! Changes the manager for a cache. All the cache entries are moved //! to the new manager, but it might not have adequate information to //! give them an accurate cost (typically applies to cost derived from //! the creation time). { Thread.MutexKey lock = cache_mgmt_mutex->lock(); // vvv Relying on the interpreter lock from here. CacheManager old_mgr = m_delete (caches, cache_name); if (old_mgr == manager) caches[cache_name] = manager; // ^^^ Relying on the interpreter lock to here. else { mapping(mixed:CacheEntry) old_lm = m_delete (old_mgr->lookup, cache_name); CacheStats old_cs = m_delete (old_mgr->stats, cache_name); // ^^^ Relying on the interpreter lock to here. old_mgr->size -= old_cs->size; cache_register (cache_name, manager); // Move over the entries. destruct (lock); int entry_size_diff = (Pike.count_memory (0, manager->CacheEntry (0, 0)) - Pike.count_memory (0, old_mgr->CacheEntry (0, 0))); foreach (old_lm; mixed key; CacheEntry old_ent) { old_mgr->remove_entry (cache_name, old_ent); CacheEntry new_ent = manager->CacheEntry (key, old_ent->data); new_ent->size = old_ent->size + entry_size_diff; manager->add_entry (cache_name, new_ent, 1, 0); } manager->update_size_limit(); // Evicts superfluous entries if necessary. } } void cache_expire (void|string cache_name) //! Expires (i.e. removes) all entries in a named cache, or in all //! caches if @[cache_name] is left out. { // Currently not very efficiently implemented, but this function // doesn't have to be quick. foreach (cache_name ? ({cache_name}) : indices (caches), string cn) { CACHE_WERR ("Emptying cache %O.\n", cn); if (CacheManager mgr = caches[cn]) { mgr->evict (0); mgr->update_size_limit(); } } } void flush_memory_cache (void|string cache_name) {cache_expire (cache_name);} void cache_clear_deltas() { cache_managers->clear_cache_context(); } mixed cache_lookup (string cache_name, mixed key, void|mapping cache_context) //! Looks up an entry in a cache. Returns @[UNDEFINED] if not found. //! //! @[cache_context] is an optional mapping used to pass info between //! @[cache_lookup] and @[cache_set], which some cache managers need //! to determine the cost of the created entry (the work done between //! a failed @[cache_lookup] and the following @[cache_set] with the //! same key is assumed to be the creation of the cache entry). //! //! If @[cache_context] is not specified, a thread local mapping is //! used. @[cache_context] is necessary when @[cache_lookup] and //! @[cache_set] are called from different threads, or in different //! callbacks from a backend. It should not be specified otherwise. //! //! If you need to use @[cache_context], create an empty mapping and //! give it to @[cache_lookup]. Then give the same mapping to the //! corresponding @[cache_set] when the entry has been created. { CacheManager mgr = caches[cache_name] || cache_register (cache_name); if (mapping(mixed:CacheEntry) lm = mgr->lookup[cache_name]) if (CacheEntry entry = lm[key]) { if (entry->timeout && entry->timeout <= time (1) || !entry->data) { mgr->remove_entry (cache_name, entry); mgr->got_miss (cache_name, key, cache_context); MORE_CACHE_WERR ("cache_lookup (%O, %s): %s\n", cache_name, RXML.utils.format_short (key), entry->data ? "Timed out" : "Destructed"); return 0; } mgr->got_hit (cache_name, entry, cache_context); MORE_CACHE_WERR ("cache_lookup (%O, %s): Hit\n", cache_name, RXML.utils.format_short (key)); return entry->data; } mgr->got_miss (cache_name, key, cache_context); MORE_CACHE_WERR ("cache_lookup (%O, %s): Miss\n", cache_name, RXML.utils.format_short (key)); return 0; } mixed cache_peek (string cache_name, mixed key) //! Checks if the cache contains an entry. Same as @[cache_lookup] //! except that it doesn't affect the hit/miss statistics or the time //! accounting used to estimate entry creation cost. { if (CacheManager mgr = caches[cache_name]) if (mapping(mixed:CacheEntry) lm = mgr->lookup[cache_name]) if (CacheEntry entry = lm[key]) { if (entry->timeout && entry->timeout <= time (1)) { mgr->remove_entry (cache_name, entry); MORE_CACHE_WERR ("cache_peek (%O, %s): Timed out\n", cache_name, RXML.utils.format_short (key)); return 0; } MORE_CACHE_WERR ("cache_peek (%O, %s): Entry found\n", cache_name, RXML.utils.format_short (key)); return entry->data; } MORE_CACHE_WERR ("cache_peek (%O, %s): Entry not found\n", cache_name, RXML.utils.format_short (key)); return 0; } mixed cache_set (string cache_name, mixed key, mixed data, void|int timeout, void|mapping|int(1..1) cache_context) //! Adds an entry to a cache. //! //! @param cache_name //! The name of the cache. The cache has preferably been created with //! @[cache_register], but otherwise it is created on-demand using the //! default cache manager. //! //! @param key //! The key for the cache entry. Normally a string, but can be //! anything that works as an index in a mapping. //! //! @param data //! The payload data. This cannot be a zero, since the cache garb will //! consider that a destructed object and evict it from the cache. //! //! @param timeout //! If nonzero, sets the maximum time in seconds that the entry is //! valid. //! //! @param cache_context //! The cache context mapping given to the earlier @[cache_lookup] //! which failed to find the entry that this call adds to the cache. //! See @[cache_lookup] for more details. //! //! @[cache_context] can also be the integer 1, which is a flag that //! this entry got created without any prior lookup, so cache managers //! tracking cost has to fall back to some default then. //! //! @returns //! Returns @[data]. //! //! @note //! Cache managers commonly uses the time from the closest preceding //! @[cache_lookup] call to calculate a weight for the entry. That //! means the caller should avoid repeated calls to @[cache_set] for //! the same entry. { ASSERT_IF_DEBUG (data); CacheManager mgr = caches[cache_name] || cache_register (cache_name); CacheEntry new_entry = mgr->CacheEntry (key, data); // We always create a new entry, even if the given key already // exists in the cache with the same data. That's to ensure we get // an up-to-date cost for the entry. (It's also a bit tricky to // atomically check for an existing entry here before creating a new // one.) #ifdef DEBUG_COUNT_MEM mapping opts = (["lookahead": DEBUG_COUNT_MEM - 1, "collect_stats": 1, "collect_direct_externals": 1, ]); float t = gauge { #else #define opts 0 #endif if (function(int|mapping:int) cm_cb = objectp (data) && data->cache_count_memory) new_entry->size = cm_cb (opts) + Pike.count_memory (-1, new_entry, key); else new_entry->size = Pike.count_memory (opts, new_entry, key, data); #ifdef DEBUG_COUNT_MEM }; werror ("%O: la %d size %d time %g int %d cyc %d ext %d vis %d revis %d " "rnd %d wqa %d\n", new_entry, opts->lookahead, opts->size, t, opts->internal, opts->cyclic, opts->external, opts->visits, opts->revisits, opts->rounds, opts->work_queue_alloc); #if 0 if (opts->external) { opts->collect_direct_externals = 1; // Raise the lookahead to 1 to recurse the closest externals. if (opts->lookahead < 1) opts->lookahead = 1; if (function(int|mapping:int) cm_cb = objectp (data) && data->cache_count_memory) res = cm_cb (opts) + Pike.count_memory (-1, entry, key); else res = Pike.count_memory (opts, entry, key, data); array exts = opts->collect_direct_externals; werror ("Externals found using lookahead %d: %O\n", opts->lookahead, exts); #if 0 foreach (exts, mixed ext) if (objectp (ext) && ext->locate_my_ext_refs) { werror ("Refs to %O:\n", ext); _locate_references (ext); } #endif } #endif #endif // DEBUG_COUNT_MEM #undef opts #ifdef DEBUG_CACHE_SIZES new_entry->cmp_size = cmp_sizeof_cache_entry (cache_name, new_entry); #endif if (timeout) new_entry->timeout = time (1) + timeout; mgr->add_entry (cache_name, new_entry, intp (cache_context) && cache_context, mappingp (cache_context) && cache_context); MORE_CACHE_WERR ("cache_set (%O, %s, %s, %O): %O\n", cache_name, RXML.utils.format_short (key), sprintf (objectp (data) ? "%O" : "%t", data), timeout, new_entry); return data; } void cache_remove (string cache_name, mixed key) //! Removes an entry from the cache. //! //! @note //! If @[key] was zero, this function used to remove the whole cache. //! Use @[cache_expire] for that instead. { MORE_CACHE_WERR ("cache_remove (%O, %O)\n", cache_name, key); if (CacheManager mgr = caches[cache_name]) if (mapping(mixed:CacheEntry) lm = mgr->lookup[cache_name]) if (CacheEntry entry = lm[key]) mgr->remove_entry (cache_name, entry); } mapping(mixed:CacheEntry) cache_entries (string cache_name) //! Returns the lookup mapping for the given named cache. Don't be //! destructive on the returned mapping or anything inside it. { if (CacheManager mgr = caches[cache_name]) if (mapping(mixed:CacheEntry) lm = mgr->lookup[cache_name]) return lm; return ([]); } array cache_indices(string|void cache_name) // Deprecated compat function. { if (!cache_name) return indices (caches); else return indices (cache_entries (cache_name)); } mapping(CacheManager:mapping(string:CacheStats)) cache_stats() //! Returns the complete cache statistics. For each cache manager, a //! mapping with the named caches it handles is returned, with their //! respective @[CacheStat] objects. Don't be destructive on any part //! of the returned value. { mapping(CacheManager:mapping(string:CacheStats)) res = ([]); foreach (cache_managers, CacheManager mgr) res[mgr] = mgr->stats; return res; } // GC statistics. These are decaying sums/averages over the last // gc_stats_period seconds. constant gc_stats_period = 60 * 60; float sum_gc_runs = 0.0, sum_gc_time = 0.0; float sum_destruct_garbage_size = 0.0; float sum_timeout_garbage_size = 0.0; float avg_destruct_garbage_ratio = 0.0; float avg_timeout_garbage_ratio = 0.0; int last_gc_run; protected void cache_clean() // Periodic gc, to clean up timed out and destructed entries. { int now = time (1); int vt = gethrvtime(), t = gethrtime(); int total_size, destr_garb_size, timeout_garb_size; CACHE_WERR ("Starting RAM cache cleanup.\n"); // Note: Might be necessary to always recheck the sizes here, since // entries can change in size for a number of reasons. Most of the // time it doesn't matter much, but the risk is that the size limit // gets unacceptably off after a while. #ifdef DEBUG_CACHE_SIZES mapping(CacheManager:int) cache_sizes = ([]); #endif foreach (caches; string cache_name; CacheManager mgr) { #ifdef DEBUG_CACHE_SIZES int cache_count, cache_size; #endif if (mapping(mixed:CacheEntry) lm = mgr->lookup[cache_name]) { foreach (lm;; CacheEntry entry) { if (!entry->data) { MORE_CACHE_WERR ("%s: Removing destructed entry %O\n", cache_name, entry); destr_garb_size += entry->size; mgr->remove_entry (cache_name, entry); } else if (entry->timeout && entry->timeout <= now) { MORE_CACHE_WERR ("%s: Removing timed out entry %O\n", cache_name, entry); timeout_garb_size += entry->size; mgr->remove_entry (cache_name, entry); } else { #ifdef DEBUG_CACHE_SIZES cache_count++; cache_size += entry->size; int size = cmp_sizeof_cache_entry (cache_name, entry); if (size != entry->cmp_size) { // Note that there are a lot of sources for false alarms here. // E.g. RXML trees can increase in size due to new <if>/<else> // branches getting visited, the entry might just happen to be in // use during the count_memory call here, and there are often // minor differences for whatever reason.. werror ("Size diff in %O: Is %d, was %d in cache_set() - " "diff %d: %O\n", cache_name, size, entry->cmp_size, size - entry->cmp_size, entry); // Update to avoid repeated messages. entry->cmp_size = size; } #endif } } } #ifdef DEBUG_CACHE_SIZES mapping(string:int)|CacheStats st = mgr->stats[cache_name] || ([]); // These might show false alarms due to races. if (cache_count != st->count) werror ("Entry count difference for %O: " "Have %d, expected %d - diff %d.\n", cache_name, cache_count, st->count, cache_count - st->count); if (cache_size != st->size) werror ("Entry size difference for %O: " "Have %d, expected %d - diff %d.\n", cache_name, cache_size, st->size, cache_size - st->size); cache_sizes[mgr] += st->size; #endif } foreach (cache_managers, CacheManager mgr) { mgr->after_gc(); total_size += mgr->size; #ifdef DEBUG_CACHE_SIZES // This might show false alarms due to races. if (cache_sizes[mgr] != mgr->size) werror ("Cache size difference for %O: " "Have %d, expected %d - diff %d.\n", mgr, cache_sizes[mgr], mgr->size, cache_sizes[mgr] - mgr->size); #endif } vt = gethrvtime() - vt; // -1 - -1 if cpu time isn't working. t = gethrtime() - t; CACHE_WERR ("Finished RAM cache cleanup: " "%db stale, %db timed out, took %s.\n", destr_garb_size, timeout_garb_size, Roxen.format_hrtime (vt || t)); int stat_last_period = now - last_gc_run; int stat_tot_period = now - cache_start_time; int startup = stat_tot_period < gc_stats_period; if (!startup) stat_tot_period = gc_stats_period; if (stat_last_period > stat_tot_period) { // GC intervals are larger than the statistics interval, so just // set the values. Note that stat_last_period is very large on the // first call since last_gc_run is zero, so we always get here then. sum_gc_time = (float) (vt || t); sum_gc_runs = 1.0; sum_destruct_garbage_size = (float) destr_garb_size; sum_timeout_garbage_size = (float) timeout_garb_size; if (total_size) { avg_destruct_garbage_ratio = (float) destr_garb_size / total_size; avg_timeout_garbage_ratio = (float) timeout_garb_size / total_size; } } else { float our_weight = (float) stat_last_period / stat_tot_period; float old_weight = 1.0 - our_weight; if (startup) { sum_gc_runs += 1.0; sum_gc_time += (float) (vt || t); sum_destruct_garbage_size += (float) destr_garb_size; sum_timeout_garbage_size += (float) timeout_garb_size; } else { sum_gc_runs = old_weight * sum_gc_runs + 1.0; sum_gc_time = old_weight * sum_gc_time + (float) (vt || t); sum_destruct_garbage_size = (old_weight * sum_destruct_garbage_size + (float) destr_garb_size); sum_timeout_garbage_size = (old_weight * sum_timeout_garbage_size + (float) timeout_garb_size); } if (total_size) { avg_destruct_garbage_ratio = (old_weight * avg_destruct_garbage_ratio + our_weight * destr_garb_size / total_size); avg_timeout_garbage_ratio = (old_weight * avg_timeout_garbage_ratio + our_weight * timeout_garb_size / total_size); } } last_gc_run = now; if (Configuration admin_config = roxenp()->get_admin_configuration()) admin_config->log_event ("roxen", "ram-cache-gc", 0, ([ "handle-cputime": vt, "handle-time": t, "stale-size": destr_garb_size, "timed-out-size": timeout_garb_size, ])); // Fall back to 60 secs just in case the config is messed up somehow. roxenp()->background_run (roxenp()->query ("mem_cache_gc_2") || 60, cache_clean); } // --- Non-garbing "cache" ----------- private mapping(string:mapping(string:mixed)) nongc_cache; //! Associates a @[value] to a @[key] in a cache identified with //! the @[cache_id]. This cache does not garb, hence it should be //! used for storing data where its size is well controled. void nongarbing_cache_set(string cache_id, string key, mixed value) { if(nongc_cache[cache_id]) nongc_cache[cache_id][key] = value; else nongc_cache[cache_id] = ([ key:value ]); } //! Returns the value associated to the @[key] in the cache //! identified by @[cache_id] in the non-garbing cache. mixed nongarbing_cache_lookup(string cache_id, string key) { return nongc_cache[cache_id]?nongc_cache[cache_id][key]:([])[0]; } //! Remove a value from the non-garbing cache. void nongarbing_cache_remove(string cache_id, string key) { if(nongc_cache[cache_id]) m_delete(nongc_cache[cache_id], key); } //! Flush a cache in the non-garbing cache. void nongarbing_cache_flush(string cache_id) { m_delete(nongc_cache, cache_id); } mapping(string:array(int)) ngc_status() { mapping(string:array(int)) res = ([]); foreach(nongc_cache; string cache; mapping(string:mixed) cachemap) { int size = Pike.count_memory (0, cachemap); res[cache] = ({ sizeof(cachemap), size}); } return res; } // --- Session cache ----------------- #ifndef SESSION_BUCKETS # define SESSION_BUCKETS 4 #endif #ifndef SESSION_SHIFT_TIME # define SESSION_SHIFT_TIME 15*60 #endif // The minimum time until which the session should be stored. private mapping(string:int) session_persistence; // The sessions, divided into several buckets. private array(mapping(string:mixed)) session_buckets; // The database for storage of the sessions. private function(string:Sql.Sql) db; // The biggest value in session_persistence private int max_persistence; // The low level call for storing a session in the database private void store_session(string id, mixed data, int t) { data = encode_value(data); db("local")->query("REPLACE INTO session_cache VALUES (%s," + t + ",%s)", id, data); } // GC that, depending on the sessions session_persistence either // throw the session away or store it in a database. private void session_cache_handler() { int t=time(1); if(max_persistence>t) { clean: foreach(session_buckets[-1]; string id; mixed data) { if(session_persistence[id]<t) { m_delete(session_buckets[-1], id); m_delete(session_persistence, id); continue; } for(int i; i<SESSION_BUCKETS-2; i++) if(session_buckets[i][id]) { continue clean; } if(objectp(data)) { m_delete(session_buckets[-1], id); m_delete(session_persistence, id); continue; } store_session(id, data, session_persistence[id]); m_delete(session_buckets[-1], id); m_delete(session_persistence, id); } } session_buckets = ({ ([]) }) + session_buckets[..SESSION_BUCKETS-2]; roxenp()->background_run(SESSION_SHIFT_TIME, session_cache_handler); } // Stores all sessions that should be persistent in the database. // This function is called upon exit. private void session_cache_destruct() { int t=time(1); if(max_persistence>t) { report_notice("Synchronizing session cache"); foreach(session_buckets, mapping(string:mixed) session_bucket) foreach(session_bucket; string id; mixed data) if(session_persistence[id]>t) { store_session(id, data, session_persistence[id]); m_delete(session_persistence, id); } } report_notice("Session cache synchronized\n"); } //! Removes the session data assiciate with @[id] from the //! session cache and session database. //! //! @seealso //! set_session_data void clear_session(string id) { m_delete(session_persistence, id); foreach(session_buckets, mapping bucket) m_delete(bucket, id); db("local")->query("DELETE FROM session_cache WHERE id=%s", id); } //! Returns the data associated with the session @[id]. //! Returns a zero type upon failure. //! //! @seealso //! set_session_data mixed get_session_data(string id) { mixed data; foreach(session_buckets, mapping bucket) if(data=bucket[id]) { session_buckets[0][id] = data; return data; } data = db("local")->query("SELECT data FROM session_cache WHERE id=%s", id); if(sizeof([array]data) && !catch(data=decode_value( ([array(mapping(string:string))]data)[0]->data ))) return data; return ([])[0]; } //! Assiciates the session @[id] to the @[data]. If no @[id] is provided //! a unique id will be generated. The session id is returned from the //! function. The minimum guaranteed storage time may be set with the //! @[persistence] argument. Note that this is a time stamp, not a time out. //! If @[store] is set, the @[data] will be stored in a database directly, //! and not when the garbage collect tries to delete the data. This //! will ensure that the data is kept safe in case the server restarts //! before the next GC. //! //! @note //! The @[data] must not contain any object, programs or functions, or the //! storage in database will throw an error. //! //! @seealso //! get_session_data, clear_session string set_session_data(mixed data, void|string id, void|int persistence, void|int(0..1) store) { if(!id) id = ([function(void:string)]roxenp()->create_unique_id)(); session_persistence[id] = persistence; session_buckets[0][id] = data; max_persistence = max(max_persistence, persistence); if(store && persistence) store_session(id, data, persistence); return id; } // Sets up the session database tables. private void setup_tables() { db("local")->query("CREATE TABLE IF NOT EXISTS session_cache (" "id CHAR(32) NOT NULL PRIMARY KEY, " "persistence INT UNSIGNED NOT NULL DEFAULT 0, " "data BLOB NOT NULL)"); master()->resolv("DBManager.is_module_table") ( 0, "local", "session_cache", "Used by the session manager" ); } //! Initializes the session handler. void init_session_cache() { db = (([function(string:function(string:object(Sql.Sql)))]master()->resolv) ("DBManager.cached_get")); setup_tables(); } void init_call_outs() { roxenp()->background_run(60, cache_clean); roxenp()->background_run (0, periodic_update_cache_size_balance); roxenp()->background_run(SESSION_SHIFT_TIME, session_cache_handler); CACHE_WERR("Cache garb call outs installed.\n"); } void create() { add_constant( "cache", this_object() ); nongc_cache = ([ ]); session_buckets = ({ ([]) }) * SESSION_BUCKETS; session_persistence = ([]); CACHE_WERR("Now online.\n"); } void destroy() { session_cache_destruct(); return; }

Roxen.git/ server/ base_server/ cache.pike > 28fdbe12ae24d05960370f8351b73a889a683ce0 [Annotate]