* or might be blanked out
* <spc><spc><spc><spc>....
*
- * site.name_duct.lock lock preventing multiple ducts
- * holder of this lock is "duct"
* F site.name main feed file
* opened/created, then written, by innd
* read by duct
* unlinked by duct
* tokens blanked out by duct when processed
- * D site.name_duct temporary feed file during flush (or crash)
+ * site.name_lock lock preventing multiple ducts
+ * to hold lock must open,F_SETLK[W]
+ * and then stat to check that locked file
+ * still has name site.name_lock
+ * holder of this lock is "duct"
+ * (only) lockholder may remove the lockfile
+ * D site.name_flushing temporary feed file during flush (or crash)
* hardlink created by duct
* unlinked by duct
- * site.name_duct.defer 431'd articles, still being written,
+ * site.name_defer 431'd articles, still being written,
* created, written, used by duct
- * site.name_backlog.lock lock taken out by innxmit wrapper
- * holder and its child are "xmit"
- * site.name_backlog_<date>.<inum>
+ *
+ * site.name_backlog.<date>.<inum>
* 431'd articles, ready for innxmit or duct
* created (link/mv) by duct
- * site.name_backlog_<anything-else> eg
- * site.name_backlog_manual
+ * site.name_backlog<anything-else> (where <anything-else> does not
+ * contain '#' or '~') eg
+ * site.name_backlog.manual
* anything the sysadmin likes (eg, feed files
* from old feeds to be merged into this one)
* created (link/mv) by admin
+ * may be symlinks (in which case links
+ * may be written through, but only links
+ * will be removed.
+ *
+ * It is safe to remove backlog files manually,
+ * if it's desired to throw away the backlog.
*
* Backlog files are also processed by innduct. We find the oldest
* backlog file which is at least a certain amount old, and feed it
void *readable_callback_user;
int fd;
- const char *path; /* ptr copy of path_<foo> or feedfile */
struct Filemon_Perfile *filemon;
oop_read *rd;
off_t offset;
Counts counts;
+ char path[];
} InputFile;
+#define SMS_LIST(X) \
+ X(WAITING) \
+ X(NORMAL) \
+ X(FLUSHING) \
+ X(FLUSHFAIL) \
+ X(SEPARATED) \
+ X(DROPPING)
+
typedef enum {
- sm_WAITING,
- sm_NORMAL,
- sm_FLUSHING,
- sm_FLUSHFAIL,
- sm_SEPARATED,
- sm_DROPPING,
+#define SMS_DEF_ENUM(s) sm_##s,
+ SMS_LIST(SMS_DEF_ENUM)
} StateMachineState;
+static const char *sms_names[]= {
+#define SMS_DEF_NAME(s) #s ,
+ SMS_LIST(SMS_DEF_NAME)
+ 0
+};
+
struct Conn {
ISNODE(Conn);
int fd, max_queue, stream;
/*----- operational variables -----*/
-static int since_connect_attempt;
static int nconns;
static LIST(Conn) idle, working, full;
static LIST(Article) *queue;
-static char *path_ductlock, *path_duct, *path_ductdefer;
+static char *path_lock, *path_flushing, *path_defer;
#define SMS(newstate, periods, why) \
(statemc_setstate(sm_##newstate,(periods),#newstate,(why)))
static StateMachineState sms;
static FILE *defer;
-static InputFile *main_input_file, *old_input_file, *backlog_input_file;
+static InputFile *main_input_file, *flushing_input_file, *backlog_input_file;
static int sm_period_counter;
logwrap(notice, "", LOG_NOTICE, -1, 0);
logwrap(info, " info", LOG_INFO, -1, 0);
+logwrap(debug, " debug", LOG_DEBUG, -1, 0);
/*========== making new connections ==========*/
static void update_nocheck(int accepted) {
accept_proportion *= accept_decay;
accept_proportion += accepted;
- nocheck= accept_proportion >= nocheck_thresh;
- if (nocheck && !nocheck_reported) {
+ int new_nocheck= accept_proportion >= nocheck_thresh;
+ if (new_nocheck && !nocheck_reported) {
notice("entering nocheck mode for the first time");
nocheck_reported= 1;
+ } else if (new_nocheck != nockech) {
+ debug("nocheck mode %s", new_nocheck ? "start" : "stop");
}
+ nocheck= new_nocheck;
}
static void article_done(Connection *conn, Article *art, int whichcount) {
open_defer();
if (fprintf(defer, "%s %s\n", TokenToText(art->token), art->messageid) <0
|| fflush(defer))
- sysfatal("write to defer file %s",path_ductdefer);
+ sysfatal("write to defer file %s",path_defer);
article_done(conn, art, RC_deferred);
break;
return;
}
- assert(ipf == old_input_file);
+ assert(ipf == flushing_input_file);
inputfile_tailing_stop(ipf);
assert(ipf->fd >= 0);
sysfatal("unable to open input file %s", path);
}
- InputFile *ipf= xmalloc(sizeof(InputFile));
+ InputFile *ipf= xmalloc(sizeof(*ipf) + strlen(path) + 1);
memset(ipf,0,sizeof(*ipf));
ipf->readable.on_readable= tailing_on_readable;
ipf->readable.try_read= tailing_try_read;
ipf->fd= fd;
- ipf->path= path;
+ strcpy(ipf->path, path);
return ipf;
}
if (ipf->fd >= 0)
if (close(ipf->fd)) sysdie("could not close input file %s", ipf->path);
- fixme maybe free ipf->path;
-
free(ipf);
}
}
ipf->offset += recsz + 1;
- if (sms==sm_NORMAL && ipf->offset >= flush_threshold) {
+ if (sms==sm_NORMAL && ipf==main_input_file &&
+ (ipf->offset >= flush_threshold || !until_spontaneous_flush) {
+
notice("starting flush (%lu >= %lu)",
(unsigned long)ipf->offset, (unsigned long)flush_threshold);
int r= link(feedfile, duct_path);
- if (r) sysdie("link feedfile %s to ductfile %s", feedfile, dut_path);
+ if (r) sysdie("link feedfile %s to flushing file %s", feedfile, dut_path);
/* => Hardlinked */
r= unlink(feedfile);
if (ipf==main_input_file) {
errno=EAGAIN;
return -1;
- } else if (ipf==old_input_file) {
+ } else if (ipf==flushing_input_file) {
assert(ipf->fd>=0);
assert(sms==sm_SEPARATED || sms==sm_DROPPING);
} else if (ipf==backlog_input_file) {
assert(!filemon_inotify_wd2ipf[wd]);
filemon_inotify_wd2ipf[wd]= ipf;
+ debug("filemon inotify startfile %p wd=%d wdmax=%d",
+ ipf, wd, filemon_inotify_wdmax);
+
pf->wd= wd;
}
static void filemon_method_stopfile(InputFile *ipf, Filemon_Perfile *pf) {
int wd= pf->wd;
+ debug("filemon inotify stopfile %p wd=%d", ipf, wd);
int r= inotify_rm_watch(filemon_inotify_fd, filemon_inotify_wd);
if (r) sysdie("inotify_rm_watch");
filemon_inotify_wd2ipf[wd]= 0;
die("inotify read %d bytes wanted struct of %d", r, (int)sizeof(iev));
}
InputFile *ipf= filemon_inotify_wd2ipf[iev.wd];
+ debug("filemon inotify readable read %p wd=%p", iev.wd, ipf);
filemon_callback(ipf);
}
return OOP_CONTINUE;
set nonblock;
loop->on_fd(loop, filemon_inotify_fd, OOP_READ, filemon_inotify_readable);
+ debug("filemon inotify init filemon_inotify_fd=%d", filemon_inotify_fd);
return 1;
}
poll for F
================
|
- | TIMEOUT
+ | TIMEOUT and no defer, no backlog
|`--------------------------.
- | | install defer as backlog
+ | |
| OPEN F SUCCEEDS | exit
,--------->| V
| V =========
| V V
| ============= ============
| SEPARATED/ DROPPING/
- | old->fd>=0 old->fd>=0
+ | flsh->fd>=0 flsh->fd>=0
| [Separated] [Dropping]
| main F idle main none
| old D tail old D tail
| V V
| =============== ===============
| SEPARATED/ DROPPING/
- | old->fd==-1 old->fd==-1
+ | flsh->fd==-1 flsh->fd==-1
| [Finishing] [Dropping]
| main F tail main none
| old D closed old D closed
struct stat stab, stabf;
int noent;
- path_ductlock= xasprintf("%s_duct.lock", feedfile);
- path_duct= xasprintf("%s_duct", feedfile);
- path_ductdefer= xasprintf("%s_duct.defer", feedfile);
- globpat_backlog= xasprintf("%s_backlog_*", feedfile);
+ path_lock= xasprintf("%s_lock", feedfile);
+ path_flushing= xasprintf("%s_flushing", feedfile);
+ path_defer= xasprintf("%s_defer", feedfile);
+ globpat_backlog= xasprintf("%s_backlog*", feedfile);
for (;;) {
- int lockfd= open(path_ductlock, O_CREAT|O_RDWR, 0600);
- if (lockfd<0) sysfatal("open lockfile %s", path_ductlock);
+ int lockfd= open(path_lock, O_CREAT|O_RDWR, 0600);
+ if (lockfd<0) sysfatal("open lockfile %s", path_lock);
struct flock fl;
memset(&fl,0,sizeof(fl));
if (quiet_if_locked) exit(0);
fatal("another duct holds the lockfile");
}
- sysdie("fcntl F_SETLK lockfile %s", path_ductlock);
+ sysdie("fcntl F_SETLK lockfile %s", path_lock);
}
xfstat_isreg(lockfd, &stabf, "lockfile");
- xlstat_isreg(path_ductlock, &stab, &noent, "lockfile");
+ xlstat_isreg(path_lock, &stab, &noent, "lockfile");
+
if (!noent && samefile(&stab, &stabf))
break;
if (close(lockfd))
- sysdie("could not close stale lockfile %s", path_ductlock);
+ sysdie("could not close stale lockfile %s", path_lock);
}
debug("startup: locked");
- xlstat_isreg(path_ductdefer, &stab, &noent, "defer file");
+ search_backlog_file();
+
+ xlstat_isreg(path_defer, &stab, &noent, "defer file");
if (noent) {
debug("startup: ductdefer ENOENT");
} else {
debug("startup: ductdefer nlink=%ld", (long)stab.st_nlink);
switch (stab.st_nlink==1) {
- case 1: /* ok */ break;
+ case 1:
+ open_defer(); /* so that we will later close it and rename it */
+ break;
case 2:
if (unlink(path_defer))
sysdie("could not unlink stale defer file link %s (presumably"
}
}
- InputFile *file_d= open_input_file(path_duct);
+ InputFile *file_d= open_input_file(path_flushing);
if (file_d) {
struct stat stab_f, stab_d;
debug("startup: F and D both exist");
- xfstat_isreg(file_d->fd, &stab_d, "ductfile");
+ xfstat_isreg(file_d->fd, &stab_d, "flushing file");
if (samefile(&stab_d, &stab_f)) {
debug("startup: F==D => Hardlinked");
- r= unlink(path_duct);
+ r= unlink(path_flushing);
if (r) sysdie("unlink feed file %s during startup", feedfile);
found_moved:
debug(" => Moved");
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);
}
if (ipf == backlog_input_file) {
notice_processed(ipf,"backlog file",ipf->path);
close_input_file(ipf);
- if (unlink(ipf->path))
- sysdie("could not unlink done backlog file %s", ipf->path);
+ 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;
}
- assert(ipf == old_input_file);
+ assert(ipf == flushing_input_file);
assert(sms==sm_SEPARATED || sms==sm_DROPPING);
notice_processed(ipf,"feed file",0);
close_defer();
- if (unlink(path_duct))
- sysdie("could not unlink old duct file %s", path_duct);
+ if (unlink(path_flushing))
+ sysdie("could not unlink old flushing file %s", path_flushing);
if (sms==sm_DROPPING) {
- notice("feed dropped and our work is complete"
- " (but check for backlog files)");
+ if (search_backlog_file()) {
+ debug("feed dropped but still backlogs to process");
+ return;
+ }
+ notice("feed dropped and our work is complete");
+ r= unlink(path_lock);
+ if (r) sysdie("unlink lockfile for old feed %s", path_lock);
exit(0);
}
open_defer();
- close_input_file(old_input_file);
- old_input_file= 0;
+ close_input_file(flushing_input_file);
+ flushing_input_file= 0;
notice("flush complete");
SMS(NORMAL, 0, "flush complete");
if (defer) return;
- defer= fopen(path_ductdefer, "a+");
- if (!defer) sysfatal("could not open defer file %s", path_ductdefer);
+ defer= fopen(path_defer, "a+");
+ if (!defer) sysfatal("could not open defer file %s", path_defer);
/* truncate away any half-written records */
xfstat_isreg(fileno(defer), &stab, "newly opened defer file");
if (stab.st_size > LONG_MAX)
- die("defer file %s size is far too large", path_ductdefer);
+ die("defer file %s size is far too large", path_defer);
if (!stab.st_size)
return;
for (;;) {
if (!truncto) break; /* was only (if anything) one half-truncated record */
if (fseek(defer, truncto-1, SEEK_SET) < 0)
- sysdie("seek in defer file %s while truncating partial", path_ductdefer);
+ sysdie("seek in defer file %s while truncating partial", path_defer);
r= getc(defer);
if (r==EOF) {
if (ferror(defer))
- sysdie("failed read from defer file %s", path_ductdefer);
+ sysdie("failed read from defer file %s", path_defer);
else
- die("defer file %s shrank while we were checking it!", path_ductdefer);
+ die("defer file %s shrank while we were checking it!", path_defer);
}
if (r=='\n') break;
truncto--;
if (stab.st_size != truncto) {
warn("truncating half-record at end of defer file %s -"
" shrinking by %ld bytes from %ld to %ld",
- path_ductdefer, orgsize - truncto, orgsize, truncto);
+ path_defer, orgsize - truncto, orgsize, truncto);
if (fflush(defer))
- sysfatal("could not flush defer file %s", path_ductdefer);
+ sysfatal("could not flush defer file %s", path_defer);
if (ftruncate(fileno(defer), truncto))
- sysdie("could not truncate defer file %s", path_ductdefer);
+ sysdie("could not truncate defer file %s", path_defer);
} else {
info("continuing existing defer file %s (%ld bytes)",
- path_ductdefer, orgsize);
+ path_defer, orgsize);
}
if (fseek(defer, truncto, SEEK_SET))
- sysdie("could not seek to new end of defer file %s", path_ductdefer);
+ sysdie("could not seek to new end of defer file %s", path_defer);
}
static void close_defer(void) {
sysdie("could not unlink old defer link %s to backlog file %s",
path_defer, backlog);
- if (backlog_nextscan_periods < 0 ||
- backlog_nextscan_periods > backlog_retry_minperiods + 1)
- backlog_nextscan_periods= backlog_retry_minperiods + 1;
+ if (until_backlog_nextscan < 0 ||
+ until_backlog_nextscan > backlog_retry_minperiods + 1)
+ until_backlog_nextscan= backlog_retry_minperiods + 1;
}
static void poll_backlog_file(void) {
- if (backlog_nextscan_periods < 0) return;
- if (backlog_nextscan_periods-- > 0) return;
+ if (until_backlog_nextscan < 0) return;
+ if (until_backlog_nextscan-- > 0) return;
search_backlog_file();
}
-static void search_backlog_file(void) {
+static int search_backlog_file(void) {
+ /* returns non-0 iff there are any backlog files */
+
glob_t gl;
int r;
struct stat stab;
const char *oldest_path=0;
time_t oldest_mtime, now;
- assert(!backlog_input_file);
+ if (backlog_input_file) return 3;
+
+ try_again:
r= glob(globpat_backlog, GLOB_ERR|GLOB_MARK|GLOB_NOSORT, 0, &gl);
case 0:
for (i=0; i<gl.gl_pathc; i++) {
const char *path= gl.gl_pathv[i];
+
+ if (strchr(path,'#') || strchr(path,'~')) {
+ debug("backlog file search skipping %s", path);
+ continue;
+ }
r= stat(path, &stab);
if (r) {
syswarn("failed to stat backlog file %s", path);
if (!oldest_path) {
debug("backlog scan: none");
- backlog_nextscan_periods= backlog_spontaneous_rescan_periods;
- return;
+ until_backlog_nextscan= backlog_spontaneous_rescan_periods;
+ return 0;
}
now= time(); if (now==-1) sysdie("time(2) failed");
debug("backlog scan: found age=%f deficiency=%ld oldest=%s",
age, age_deficiency, oldest_path);
- backlog_input_file= open_input_file();
+ backlog_input_file= open_input_file(oldest_path);
+ if (!backlog_input_file) {
+ warn("backlog file %s vanished as we opened it", backlog_input_file);
+ goto try_again;
+ }
inputfile_tailing_start(backlog_input_file);
- backlog_nextscan_periods= -1;
- return;
+ until_backlog_nextscan= -1;
+ return 1;
}
- backlog_nextscan_periods= age_deficiency / PERIOD_SECONDS;
+ until_backlog_nextscan= age_deficiency / PERIOD_SECONDS;
if (backlog_spontaneous_rescan_periods >= 0 &&
- backlog_nextscan_periods > backlog_spontaneous_rescan_periods)
- backlog_nextscan_periods= backlog_spontaneous_rescan_periods;
+ until_backlog_nextscan > backlog_spontaneous_rescan_periods)
+ until_backlog_nextscan= backlog_spontaneous_rescan_periods;
debug("backlog scan: young age=%f deficiency=%ld nextscan=%d oldest=%s",
- age, age_deficiency, backlog_nextscan_periods, oldest_path);
+ age, age_deficiency, until_backlog_nextscan, oldest_path);
+ return 2;
}
/*========== flushing the feed ==========*/
loop->cancel_fd(fd);
close(fd);
- assert(!old_input_file);
+ assert(!flushing_input_file);
if (WIFEXITED(status)) {
switch (WEXITSTATUS(status)) {
case INNDCOMMCHILD_ESTATUS_NONESUCH:
warn("feed has been dropped by innd, finishing up");
- old_input_file= main_input_file;
+ flushing_input_file= main_input_file;
main_input_file= 0;
SMS(DROPPING, 0, "dropped by innd");
return OOP_CONTINUE;
case 0:
- old_input_file= main_input_file;
+ flushing_input_file= main_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");
return OOP_CONTINUE;
sysdie("%s child: failed to reset SIGPIPE");
postfork_inputfile(main_input_file);
- postfork_inputfile(old_input_file);
+ postfork_inputfile(flushing_input_file);
postfork_conns(idle.head);
postfork_conns(working.head);
postfork_conns(full.head);
if (main_input_file && main_input_file->readable_callback)
filemon_callback(main_input_file);
});
+
+#define DEBUGF_IPF(wh) " " #wh "=%p/%s:ip=%ld,off=%ld,fd=%d%s" \
+#define DEBUG_IPF(sh) \
+ wh##_input_file, debug_ipf_path(wh##_input_file), \
+ wh##_input_file->inprogress, (long)wh##_input_file->offset, \
+ wh##_input_file->fd, wh##_input_file->rd ? "+" : ""
+static const char *debug_ipf_path(InputFile *ipf) {
+ char *slash= strrchr(ipf->path,'/');
+ return slash ? slash+1 : ipf->path;
+}
EVERY(period, {PERIOD_SECONDS,0}, {
+ debug("PERIOD"
+ " sms=%s queue=%d sm_period_counter=%d"
+ " connect_delay=%d until_spontaneous_flush=%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,
+ 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();
else if (feedfile[strlen(feedfile)-1]=='/')
feedfile= xasprintf("%s%s",feedfile,sitename);
- const char *feedfile_forbidden= "?*[";
+ const char *feedfile_forbidden= "?*[~#";
int c;
while ((c= *feedfile_forbidden++))
if (strchr(feedfile, c))
- badusage("feed filename may not contain glob metacharacter %c",c);
+ badusage("feed filename may not contain metacharacter %c",c);
if (signal(SIGPIPE, SIG_IGN) == SIG_ERR)
sysdie("could not ignore SIGPIPE");
~ian / inn-innduct.git / blobdiff