X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=innduct.git;a=blobdiff_plain;f=backends%2Finnduct.c;h=b9460ed598a62a205ef865608cbab2151815bc49;hp=b360d9540d10dc310db146ed370d0b9ce3c44a94;hb=2092e752db3c70a7f576390ce8744d0618df088b;hpb=84b25112929fba84e4af13907b0d30209deaf404 diff --git a/backends/innduct.c b/backends/innduct.c index b360d95..b9460ed 100644 --- a/backends/innduct.c +++ b/backends/innduct.c @@ -56,6 +56,7 @@ * If we don't have a backlog file that we're reading, we close the * defer file that we're writing and make it into a backlog file at * the first convenient opportunity. + * -8<- OVERALL STATES: @@ -135,7 +136,12 @@ "duct reading" means innduct is reading the file but also overwriting processed tokens. + * ->8- -^L- * + * rune for printing diagrams: + +perl -ne 'print if m/-8\<-/..m/-\>8-/; print "\f" if m/-\^L-/' backends/innduct.c |a2ps -R -B -ops + * */ @@ -260,12 +266,12 @@ typedef struct InputFile { } InputFile; #define SMS_LIST(X) \ - X(WAITING) \ X(NORMAL) \ X(FLUSHING) \ - X(FLUSHFAIL) \ + X(FLUSHFAILED) \ X(SEPARATED) \ - X(DROPPING) + X(DROPPING) \ + X(DROPPED) typedef enum { #define SMS_DEF_ENUM(s) sm_##s, @@ -981,8 +987,8 @@ static void article_done(Connection *conn, Article *art, int whichcount) { ipf->inprogress--; assert(ipf->inprogress >= 0); - if (!ipf->inprogress) - loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, ipf); + if (!ipf->inprogress && ipf != main_input_file) + queue_check_input_done(); free(art); } @@ -1092,26 +1098,20 @@ static void *peer_rd_ok(oop_source *lp, oop_read *oread, oop_event ev, static void feedfile_eof(InputFile *ipf) { assert(ipf != main_input_file); /* promised by tailing_try_read */ - inputfile_tailing_stop(ipf); - - if (ipf == backlog_input_file) { - assert(ipf->fd >= 0); - if (close(ipf->fd)) sysdie("could not close backlog file %s", ipf->path); - ipf->fd= -1; - return; - } - - assert(ipf == flushing_input_file); - inputfile_tailing_stop(ipf); assert(ipf->fd >= 0); if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path); ipf->fd= -1; - assert(sms==sm_SEPARATED || sms==sm_DROPPING); - - if (main_input_file) - inputfile_tailing_start(main_input_file); + if (ipf == flushing_input_file) { + assert(sms==sm_SEPARATED || sms==sm_DROPPING); + if (main_input_file) inputfile_tailing_start(main_input_file); + statemc_check_flushing_done(); + } else if (ipf == backlog_input_file) { + statemc_check_backlog_done(); + } else { + abort(); /* supposed to wait rather than get EOF on main input file */ + } } static InputFile *open_input_file(const char *path) { @@ -1142,8 +1142,6 @@ static void close_input_file(InputFile *ipf) { if (ipf->fd >= 0) if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path); - - free(ipf); } @@ -1184,25 +1182,31 @@ typedef void *feedfile_got_article(oop_source *lp, oop_read *rd, ipf->offset += recsz + 1; if (sms==sm_NORMAL && ipf==main_input_file && - (ipf->offset >= flush_threshold || !until_spontaneous_flush) { + ipf->offset >= flush_threshold) + statemc_start_flush("feed file size"); + + check_master_queue(); +} - notice("starting flush (%lu >= %lu)", - (unsigned long)ipf->offset, (unsigned long)flush_threshold); +static void statemc_start_flush(const char *why) { /* Normal => Flushing */ + assert(sms == sm_NORMAL); - int r= link(feedfile, duct_path); - if (r) sysdie("link feedfile %s to flushing file %s", feedfile, dut_path); - /* => Hardlinked */ + debug("starting flush (%s) (%lu >= %lu) (%d)", + why, + (unsigned long)ipf->offset, (unsigned long)flush_threshold, + sm_period_counter); - xunlink(feedfile, "old feedfile link"); - /* => Moved */ + int r= link(feedfile, duct_path); + if (r) sysdie("link feedfile %s to flushing file %s", feedfile, + path_duct); + /* => Hardlinked */ - spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */ - } + xunlink(feedfile, "old feedfile link"); + /* => Moved */ - check_master_queue(); + spawn_inndcomm_flush(why); /* => Flushing FLUSHING */ } - /*========== tailing input file ==========*/ static void filemon_start(InputFile *ipf) { @@ -1238,6 +1242,12 @@ static void on_cancel(struct oop_readable *rable) { ipf->readable_callback= 0; } +static void tailing_queue_readable(InputFile *ipf) { + /* lifetime of ipf here is OK because destruction will cause + * on_cancel which will cancel this callback */ + loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf); +} + static int tailing_on_readable(struct oop_readable *rable, oop_readable_call *cb, void *user) { InputFile *ipf= (void*)rable; @@ -1247,7 +1257,7 @@ static int tailing_on_readable(struct oop_readable *rable, ipf->readable_callback_user= user; filemon_startfile(ipf); - loop->on_time(loop, OOP_TIME_NOW, tailing_rable_call_time, ipf); + tailing_queue_readable(ipf); return 0; } @@ -1409,6 +1419,7 @@ static void inputfile_tailing_stop(InputFile *ipf) { /* See official state diagram at top of file. We implement * this as follows: + * -8<- .=======. ||START|| @@ -1441,63 +1452,72 @@ static void inputfile_tailing_stop(InputFile *ipf) { | | | | | spawn inndcomm flush | | V | - | ================ | - | FLUSHING | + | ================== | + | FLUSHING[-ABSENT] | | [Flushing] | | main D tail/none | - | ================ | + | ================== | | | | | | INNDCOMM FLUSH FAILS ^ | |`----------------------->----------. | | | | | | | NO SUCH SITE V | - ^ |`--------------->----. ================ | - | | \ FLUSHFAIL | + ^ |`--------------->----. ==================== | + | | \ FLUSHFAILED[-ABSENT] | | | \ [Moved] | | | FLUSH OK \ main D tail/none | - | | open F \ ================ | + | | open F \ ==================== | | | \ | | | | \ | TIME TO RETRY | - |<--------'| ,---<---'\ `----------------' - | D NONE | | D NONE \ - | | | \ - | | | | - | | V | - | V | V - | ============= | ============ - | SEPARATED/ | DROPPING/ - | flsh->fd>=0 | flsh->fd>=0 - | [Separated] | [Dropping] - | main F idle | main none - | old D tail | old D tail - | ============= | ============ - | | | | - ^ | EOF ON D | | EOF ON D - | V | V - | =============== | =============== - | SEPARATED/ | DROPPING/ - | flsh->fd==-1 V flsh->fd==-1 - | [Finishing] | [Dropping] - | main F tail `. main none - | old D closed `. old D closed - | =============== `. =============== - | | `. | - | | ALL D PROCESSED `. | ALL D PROCESSED - | V install defer as backlog `. V install defer as backlog - ^ | close D \| close D - | | unlink D | unlink D - | | | unlink lock - | | | exit - `----------' V - ========== - (ESRCH) - [Droppped] - ========== + | |`------->----. ,---<---'\ `----------------' + | | D NONE | | D NONE `----. + | V | | V + | ============= V V ============ + | SEPARATED-1 | | DROPPING-1 + | flsh->fd>=0 | | flsh->fd>=0 + | [Separated] | | [Dropping] + | main F idle | | main none + | old D tail | | old D tail + | ============= | | ============ + | | | | install | + ^ | EOF ON D | | defer | EOF ON D + | V | | V + | =============== | | =============== + | SEPARATED-2 | | DROPPING-2 + | flsh->fd==-1 | V flsh->fd==-1 + | [Finishing] | | [Dropping] + | main F tail | `. main none + | old D closed | `. old D closed + | =============== V `. =============== + | | `. | + | | ALL D PROCESSED `. | ALL D PROCESSED + | V install defer as backlog `. | install defer + ^ | close D `. | close D + | | unlink D `. | unlink D + | | | | + | | V V + `----------' ============== + DROPPED + [Dropped] + main none + old none + some backlog + ============== + | + | ALL BACKLOG DONE + | + | unlink lock + | exit + V + ========== + (ESRCH) + [Droppped] + ========== + * ->8- */ static void statemc_init(void) { struct stat stab, stabf; - int noent; path_lock= xasprintf("%s_lock", feedfile); path_flushing= xasprintf("%s_flushing", feedfile); @@ -1522,9 +1542,10 @@ static void statemc_init(void) { } xfstat_isreg(lockfd, &stabf, "lockfile"); - xlstat_isreg(path_lock, &stab, &noent, "lockfile"); + int lock_noent; + xlstat_isreg(path_lock, &stab, &lock_noent, "lockfile"); - if (!noent && samefile(&stab, &stabf)) + if (!lock_noent && samefile(&stab, &stabf)) break; if (close(lockfd)) @@ -1534,8 +1555,9 @@ static void statemc_init(void) { search_backlog_file(); - xlstat_isreg(path_defer, &stab, &noent, "defer file"); - if (noent) { + int defer_noent; + xlstat_isreg(path_defer, &stab, &defer_noent, "defer file"); + if (defer_noent) { debug("startup: ductdefer ENOENT"); } else { debug("startup: ductdefer nlink=%ld", (long)stab.st_nlink); @@ -1553,96 +1575,94 @@ static void statemc_init(void) { } } - InputFile *file_d= open_input_file(path_flushing); - - if (file_d) { - struct stat stab_f, stab_d; + struct stat stab_f, stab_d; + int noent_f; - xlstat_isreg(feedfile, &stab_f, &noent, "feed file"); - if (noent) { - debug("startup: D exists, F ENOENT => Moved"); - goto found_moved; - } + InputFile *file_d= open_input_file(path_flushing); + if (file_d) xfstat_isreg(file_d->fd, &stab_d, "flushing file"); - debug("startup: F and D both exist"); + xlstat_isreg(feedfile, &stab_f, &noent_f, "feedfile"); - xfstat_isreg(file_d->fd, &stab_d, "flushing file"); + if (!noent_f && file_d && samefile(&stab_f, &stab_d)) { + debug("startup: F==D => Hardlinked"); + xunlink(feedfile, "feed file (during startup)"); /* => Moved */ + noent_f= 1; + } - if (samefile(&stab_d, &stab_f)) { - debug("startup: F==D => Hardlinked"); - xunlink(path_flushing, "feed file (during startup)"); - found_moved: - debug(" => Moved"); + if (noent_f) { + debug("startup: F ENOENT => Moved"); + if (file_d) startup_set_input_file(file_d); + spawn_inndcomm_flush("feedfile missing at startup"); + /* => Flushing, sms:=FLUSHING */ + } else { + if (file_d) { + debug("startup: F!=D => Separated"); startup_set_input_file(file_d); - spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */ - } else { - debug("F!=D => Separated"); SMS(SEPARATED, 0, "found both old and current feed files"); - startup_set_input_file(file_d); + } else { + debug("startup: F exists, D ENOENT => Normal"); + FILE *file_f= open_input_file(feedfile); + if (!file_f) die("feed file vanished during startup"); + startup_set_input_file(file_f); + SMS(NORMAL, flushfail_retry_periods, "normal startup"); } - } else { - debug("startup: D ENOENT => Nothing"); - SMS(WAITING, open_wait_periods, "no feed file currently exists"); } } -static void statemc_poll(void) { - if (sms==sm_WAITING) { statemc_waiting_poll(); return; } - +static void statemc_period_poll(void) { if (!sm_period_counter) return; sm_period_counter--; assert(sm_period_counter>=0); if (sm_period_counter) return; switch (sms) { - case sm_WAITING: - fatal("timed out waiting for innd to create feed file %s", feedfile); - case sm_FLUSHFAIL: - spawn_inndcomm_flush(void); + case sm_NORMAL: + statemc_start_flush("periodic"); /* Normal => Flushing; => FLUSHING */ + break; + case sm_FLUSHFAILED: + spawn_inndcomm_flush("retry"); /* Moved => Flushing; => FLUSHING */ break; default: abort(); } } -static void statemc_waiting_poll(void) { - InputFile *file_f= open_input_file(feedfile); - if (!file_f) return; - startup_set_input_file(file_d); - SMS(NORMAL, 0, "found and opened feed file"); -} - static void startup_set_input_file(InputFile *f) { assert(!main_input_file); main_input_file= f; - until_spontaneous_flush= spontaneous_flush_periods; inputfile_tailing_start(f); } -static void *statemc_check_input_done(oop_source *lp, - struct timeval now, void *ipf_v) { - InputFile *ipf= ipf_v; - struct stat stab; - - if (ipf->inprogress) return; /* new article in the meantime */ - if (ipf->fd >= 0); return; /* not had EOF */ - - if (ipf == backlog_input_file) { - notice_processed(ipf,"backlog file",ipf->path); - close_input_file(ipf); - if (unlink(ipf->path)) { - if (errno != ENOENT) - sysdie("could not unlink processed backlog file %s", ipf->path); - warn("backlog file %s vanished while we were reading it" - " so we couldn't remove it (but it's done now, anyway)", - ipf->path); - } - backlog_input_file= 0; - search_backlog_file(); - return; +static int inputfile_is_done(InputFile *ipf) { + if (!ipf) return 0; + if (ipf->inprogress) return 0; /* new article in the meantime */ + if (ipf->fd >= 0); return 0; /* not had EOF */ + return 1; +} + +static void statemc_check_backlog_done(void) { + InputFile *ipf= backlog_input_file(); + if (!inputfile_is_done(ipf)) return; + + notice_processed(ipf,"backlog file",ipf->path); + close_input_file(ipf); + if (unlink(ipf->path)) { + if (errno != ENOENT) + sysdie("could not unlink processed backlog file %s", ipf->path); + warn("backlog file %s vanished while we were reading it" + " so we couldn't remove it (but it's done now, anyway)", + ipf->path); } + free(ipf); + backlog_input_file= 0; + search_backlog_file(); + return; +} + +static void statemc_check_flushing_done(void) { + InputFile *ipf= flushing_input_file; + if (!inputfile_is_done(ipf)) return; - assert(ipf == flushing_input_file); assert(sms==sm_SEPARATED || sms==sm_DROPPING); notice_processed(ipf,"feed file",0); @@ -1651,33 +1671,52 @@ static void *statemc_check_input_done(oop_source *lp, xunlink(path_flushing, "old flushing file"); - if (sms==sm_DROPPING) { - if (search_backlog_file()) { - debug("feed dropped but still backlogs to process"); - return; - } - notice("feed dropped and our work is complete"); - xunlink(path_lock, "lockfile for old feed"); - exit(0); - } - - open_defer(); - close_input_file(flushing_input_file); + free(flushing_input_file); flushing_input_file= 0; - notice("flush complete"); - SMS(NORMAL, 0, "flush complete"); + if (sms==sm_SEPARATED) { + notice("flush complete"); + SMS(NORMAL, 0, "flush complete"); + } else if (sms==sm_DROPPING) { + SMS(DROPPED, 0, "old flush complete"); + search_backlog_file(); + notice("feed dropped, but will continue until backlog is finished"); + } +} + +static void *statemc_check_input_done(oop_source *lp, struct timeval now, + void *u) { + assert(!inputfile_is_done(main_input_file)); + statemc_check_flushing_done(); + statemc_check_backlog_done(); + return OOP_CONTINUE; +} + +static void queue_check_input_done(void) { + loop->on_time(loop, OOP_TIME_NOW, statemc_check_input_done, 0); } static void statemc_setstate(StateMachineState newsms, int periods, const char *forlog, const char *why) { sms= newsms; sm_period_counter= periods; + + const char *xtra= ""; + switch (sms) { + case sm_FLUSHING: sm_FLUSHFAILED: + if (!main_input_file) xtra= "-ABSENT"; + break; + case sm_SEPARATED: case sm_DROPPING: + xtra= flushing_input_file->fd >= 0 ? "-1" : "-2"; + break; + default:; + } + if (periods) { - info("%s[%d] %s",periods,forlog,why); + info("%s%s[%d] %s",forlog,xtra,periods,why); } else { - info("%s %s",forlog,why); + info("%s%s %s",forlog,xtra,why); } } @@ -1767,7 +1806,7 @@ static void poll_backlog_file(void) { search_backlog_file(); } -static int search_backlog_file(void) { +static void search_backlog_file(void) { /* returns non-0 iff there are any backlog files */ glob_t gl; @@ -1820,6 +1859,12 @@ static int search_backlog_file(void) { if (!oldest_path) { debug("backlog scan: none"); + + if (sms==sm_DROPPED) { + notice("feed dropped and our work is complete"); + xunlink(path_lock, "lockfile for old feed"); + exit(0); + } until_backlog_nextscan= backlog_spontaneous_rescan_periods; return 0; } @@ -1874,17 +1919,37 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { case INNDCOMMCHILD_ESTATUS_NONESUCH: warn("feed has been dropped by innd, finishing up"); flushing_input_file= main_input_file; + tailing_queue_readable(flushing_input_file); + /* we probably previously returned EAGAIN from our fake read method + * when in fact we were at EOF, so signal another readable event + * so we actually see the EOF */ + main_input_file= 0; - SMS(DROPPING, 0, "dropped by innd"); + + if (flushing_input_file) { + SMS(DROPPING, 0, "feed dropped by innd, but must finish last flush"); + } else { + close_defer(); + SMS(DROPPED, 0, "feed dropped by innd"); + search_backlog_file(); + } return OOP_CONTINUE; case 0: + /* as above */ flushing_input_file= main_input_file; + tailing_queue_readable(flushing_input_file); + main_input_file= open_input_file(feedfile); if (!main_input_file) die("flush succeeded but feedfile %s does not exist!", feedfile); - until_spontaneous_flush= spontaneous_flush_periods; - SMS(SEPARATED, 0, "feed file missing"); + + if (flushing_input_file) { + SMS(SEPARATED, spontaneous_flush_periods, "recovery flush complete"); + } else { + close_defer(); + SMS(NORMAL, spontaneous_flush_periods, "flush complete"); + } return OOP_CONTINUE; default: @@ -1900,7 +1965,7 @@ static void *inndcomm_event(oop_source *lp, int fd, oop_event e, void *u) { } failed: - SMS(FLUSHFAIL, flushfail_retry_periods, "flush failed, will retry"); + SMS(FLUSHFAILED, flushfail_retry_periods, "flush failed, will retry"); } static void inndcommfail(const char *what) { @@ -1908,10 +1973,12 @@ static void inndcommfail(const char *what) { exit(INNDCOMMCHILD_ESTATUS_FAIL); } -void spawn_inndcomm_flush(void) { +void spawn_inndcomm_flush(const char *why) { /* Moved => Flushing */ int pipefds[2]; - assert(sms==sm_NORMAL || sms==sm_FLUSHFAIL); + notice("flushing %s",why); + + assert(sms==sm_NORMAL || sms==sm_FLUSHFAILED); assert(!inndcomm_child); if (pipe(pipefds)) sysdie("create pipe for inndcomm child sentinel"); @@ -1938,7 +2005,7 @@ void spawn_inndcomm_flush(void) { int sentinel_fd= pipefds[0]; on_fd_read_except(sentinel_fd, inndcomm_event); - SMS(FLUSHING, 0, "flush is in progress"); + SMS(FLUSHING, 0, why); } /*========== main program ==========*/ @@ -2002,23 +2069,22 @@ static const char *debug_ipf_path(InputFile *ipf) { EVERY(period, {PERIOD_SECONDS,0}, { debug("PERIOD" - " sms=%s queue=%d sm_period_counter=%d" - " connect_delay=%d until_spontaneous_flush=%d" + " sms=%s[%d] queue=%d connect_delay=%d" " input_files" DEBUGF_IPF(main) DEBUGF_IPF(old) DEBUGF_FMT(flushing) " conns idle=%d working=%d full=%d" " children connecting=%ld inndcomm_child" , - sms_names[sms], queue.count, sm_period_counter, - connect_delay, until_spontaneous_flush, + sms_names[sms], sm_period_counter, queue.count, connect_delay, DEBUG_IPF(main), DEBUG_IPF(flushing), DEBUG_IPF(flushing), idle.count, working.count, full.count, (long)connecting_child, (long)inndcomm_child ); + if (connect_delay) connect_delay--; - if (until_spontaneous_flush) until_spontaneous_flush--; + poll_backlog_file(); if (!backlog_input_file) close_defer(); /* want to start on a new backlog */ - statemc_poll(); + statemc_period_poll(); check_master_queue(); }); @@ -2120,7 +2186,6 @@ static const Option options[]= { { 0, "max-queue-size", &max_queue_per_conn op_integer }, { 0, "reconnect-interval", &reconnect_delay_periods, op_periods_rndup }, { 0, "flush-retry-interval", &flushfail_retry_periods, op_periods_rndup }, -{ 0, "feedfile-open-timeout", &open_wait_periods, op_periods_rndup }, { 0, "connection-timeout", &connection_timeout, op_seconds }, { 0, "inndcomm-timeout", &inndcomm_flush_timeout, op_seconds }, };