START
|
- check D, F
- |
- <--------------------------------------'|
- Nothing F, D both ENOENT |
- F: ENOENT |
- D: ENOENT |
- duct: not not reading anything |
- | |
- |`---------------------. |
- | | duct times out waiting for F |
- V innd creates F | duct exits |
- | V |
- Noduct GO TO Dropped |
- F: innd writing |
- D: ENOENT |
- duct: not running or not reading anything |
- | |
- | |
- ,-->--+ <---------------------------------'|
- | | duct opens F F exists |
- | | D ENOENT |
+ ,-->--. check F, D
+ | | |
+ | | |
+ | | <----------------<---------------------------------'|
+ | | F exists |
+ | | D ENOENT |
+ | | duct opens F |
| V |
| Normal |
| F: innd writing, duct reading |
| F == D: innd writing, duct reading both exist |
^ | |
| | duct unlinks F |
- | V |
- | Moved <----+------------<--'|
- | F: ENOENT | F ENOENT |
- | D: innd writing, duct reading | D exists |
- | | | |
- | | duct requests flush of feed | |
- | | (others can too, harmlessly) | |
- | V | |
- | Flushing | |
- | F: ENOENT | |
- | D: innd flushing, duct reading | |
- | | | |
- | | inndcomm flush fails | |
- | |`-------------------------->---------' |
- | | |
+ | | <-----------<-------------<--'|
+ | | open D F ENOENT |
+ | | if exists |
+ | | |
+ | V <---------------------. |
+ | Moved | |
+ | F: ENOENT | |
+ | D: innd writing, duct reading; or ENOENT | |
+ | | | |
+ | | duct requests flush of feed | |
+ | | (others can too, harmlessly) | |
+ | V | |
+ | Flushing | |
+ | F: ENOENT | |
+ | D: innd flushing, duct; or ENOENT | |
+ | | | |
+ | | inndcomm flush fails | |
+ | |`-------------------------->------------------' |
+ | | |
| | inndcomm reports no such site |
| |`---------------------------------------------------- | -.
- | | | |
+ | | | |
| | innd finishes writing D, creates F | |
| | inndcomm reports flush successful | |
| | | |
| V | |
| Separated <----------------' |
| F: innd writing F!=D /
- | D: duct reading both exist /
+ | D: duct reading; or ENOENT both exist /
| | /
| | duct gets to the end of D /
| | duct opens F too /
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 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;
return status;
}
+static void xunlink(const char *path, const char *what) {
+ int r= unlink(path);
+ if (r) sysdie("can't unlink %s %s", path, what);
+}
+
static void check_isreg(const struct stat *stab, const char *path,
const char *what) {
if (!S_ISREG(stab->st_mode))
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) {
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);
if (r) sysdie("link feedfile %s to flushing file %s", feedfile, dut_path);
/* => Hardlinked */
- r= unlink(feedfile);
- if (r) sysdie("unlink old feedfile link %s", feedfile);
+ xunlink(feedfile, "old feedfile link");
/* => Moved */
spawn_inndcomm_flush(); /* => Flushing, sets sms to sm_FLUSHING */
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;
}
/* See official state diagram at top of file. We implement
* this as follows:
- ================
- WAITING
- [Nothing/Noduct]
- poll for F
- ================
- |
- | TIMEOUT and no defer, no backlog
- |`--------------------------.
- | |
- | OPEN F SUCCEEDS | exit
- ,--------->| V
- | V =========
- | ======== (ESRCH)
- | NORMAL [Dropped]
- | [Normal] =========
- | main F tail
- | ========
- | |
- | | F IS SO BIG WE SHOULD FLUSH
- ^ | hardlink F to D
- | [Hardlinked]
- | | unlink F
- | | our handle onto F is now onto D
- | [Moved]
- | |
- | |<---------------------------------------------------.
+ .=======.
+ ||START||
+ `======='
+ |
+ | open F
+ |
+ | F ENOENT
+ |`---------------------------------------------------.
+ F OPEN OK | |
+ |`---------------- - - - |
+ D ENOENT | D EXISTS see OVERALL STATES diagram |
+ | for full startup logic |
+ ,--------->| |
+ | V |
+ | ============ try to |
+ | NORMAL open D |
+ | [Normal] |
+ | main F tail |
+ | ============ V
+ | | |
+ | | F IS SO BIG WE SHOULD FLUSH, OR TIMEOUT |
+ ^ | hardlink F to D |
+ | [Hardlinked] |
+ | | unlink F |
+ | | our handle onto F is now onto D |
+ | [Moved] |
+ | | |
+ | |<-------------------<---------------------<---------+
| | |
| | spawn inndcomm flush |
| V |
- | ========== |
+ | ================ |
| FLUSHING |
| [Flushing] |
- | main D tail |
- | ========== |
+ | main D tail/none |
+ | ================ |
| | |
| | INNDCOMM FLUSH FAILS ^
| |`----------------------->----------. |
| | | |
| | NO SUCH SITE V |
- ^ |`--------------->----. =========== |
- | | \ FLUSHFAIL |
- | | \ [Moved] |
- | | \ main D tail |
- | | \ =========== |
+ ^ |`--------------->----. ================ |
+ | | \ FLUSHFAIL |
+ | | \ [Moved] |
+ | | FLUSH OK \ main D tail/none |
+ | | open F \ ================ |
| | \ | |
| | \ | TIME TO RETRY |
- | | \ `----------------'
- | | FLUSH OK \
- | | open F \
- | V V
- | ============= ============
- | SEPARATED/ DROPPING/
- | old->fd>=0 old->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/
- | old->fd==-1 old->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
+ |<--------'| ,---<---'\ `----------------'
+ | 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
==========
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"
- " hardlink to backlog file)", path_defer);
+ xunlink(path_defer, "stale defer file link"
+ " (presumably hardlink to backlog file)");
break;
default:
die("defer file %s has unexpected link count %d",
if (samefile(&stab_d, &stab_f)) {
debug("startup: F==D => Hardlinked");
- r= unlink(path_flushing);
- if (r) sysdie("unlink feed file %s during startup", feedfile);
+ xunlink(path_flushing, "feed file (during startup)");
found_moved:
debug(" => Moved");
startup_set_input_file(file_d);
}
} else {
debug("startup: D ENOENT => Nothing");
- fixme need to try flushing innd here - needs state diagram changes;
SMS(WAITING, open_wait_periods, "no feed file currently exists");
}
}
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);
}
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_flushing))
- sysdie("could not unlink old flushing file %s", path_flushing);
+ xunlink(path_flushing, "old flushing file");
if (sms==sm_DROPPING) {
if (search_backlog_file()) {
return;
}
notice("feed dropped and our work is complete");
- r= unlink(path_lock);
- if (r) sysdie("unlink lockfile for old feed %s", path_lock);
+ xunlink(path_lock, "lockfile for old feed");
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");
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();
}
if (!oldest_path) {
debug("backlog scan: none");
- backlog_nextscan_periods= backlog_spontaneous_rescan_periods;
+ until_backlog_nextscan= backlog_spontaneous_rescan_periods;
return 0;
}
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;
+ 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;
}
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);
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();
}
if (!arg) badusage("need site name argument");
- if (*++argv) badusage("too many non-option arguments");
sitename= arg;
+ if ((arg= *++argv)) {
+ if (remote_host) badusage("must not specify -h and also host as argument");
+ remote_host= arg;
+ }
+
+ if (*++argv) badusage("too many non-option arguments");
+
if (nocheck_thresh_pct < 0 || nocheck_thresh_pct > 100)
badusage("nocheck threshold percentage must be between 0..100");
nocheck_thresh= nocheck_thresh_pct * 0.01;
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");