static int update_jcb_obj (flux_t *h, int64_t jobid, const char *key,
json_object *jcb)
{
int rc = -1;
json_object *o = NULL;
if (!jcb) return -1;
if (jobid_exist (h, jobid) != 0) return -1;
if (is_jobid (key)) {
flux_log (h, LOG_ERR, "jobid attr cannot be updated");
} else if (is_state_pair (key)) {
if (Jget_obj (jcb, JSC_STATE_PAIR, &o))
rc = update_state (h, jobid, o);
} else if (is_rdesc (key)) {
if (Jget_obj (jcb, JSC_RDESC, &o))
rc = update_rdesc (h, jobid, o);
} else if (is_rdl (key)) {
const char *s = NULL;
if (Jget_str (jcb, JSC_RDL, &s))
rc = update_rdl (h, jobid, s);
} else if (is_rdl_alloc (key)) {
if (Jget_obj (jcb, JSC_RDL_ALLOC, &o))
rc = update_rdl_alloc (h, jobid, o);
} else if (is_pdesc (key)) {
if (Jget_obj (jcb, JSC_PDESC, &o))
rc = update_pdesc (h, jobid, o);
}
else
flux_log (h, LOG_ERR, "key (%s) not understood", key);
return rc;
}
static int waitjob_cb (const char *jcbstr, void *arg, int errnum)
{
JSON jcb = NULL;
int64_t os = 0, ns = 0, j = 0;
flux_t h = (flux_t)arg;
wjctx_t *ctx = getctx (h);
if (errnum > 0) {
flux_log (ctx->h, LOG_ERR, "waitjob_cb: errnum passed in");
return -1;
}
if (!(jcb = Jfromstr (jcbstr))) {
flux_log (ctx->h, LOG_ERR, "waitjob_cb: error parsing JSON string");
return -1;
}
get_jobid (jcb, &j);
get_states (jcb, &os, &ns);
Jput (jcb);
if ((j == ctx->jobid) && (ns == J_COMPLETE)) {
if (ctx->complete)
touch_outfile (ctx->complete);
flux_log (ctx->h, LOG_INFO, "waitjob_cb: completion notified");
raise (SIGINT);
}
return 0;
}
static void fixup_newjob_event (flux_t *h, int64_t nj)
{
json_object *ss = NULL;
json_object *jcb = NULL;
int64_t js = J_NULL;
char *key = xasprintf ("%"PRId64, nj);
jscctx_t *ctx = getctx (h);
/* We fix up ordering problem only when new job
event hasn't been reported through a kvs watch
*/
jcb = Jnew ();
ss = Jnew ();
Jadd_int64 (jcb, JSC_JOBID, nj);
Jadd_int64 (ss, JSC_STATE_PAIR_OSTATE , (int64_t) js);
Jadd_int64 (ss, JSC_STATE_PAIR_NSTATE, (int64_t) js);
json_object_object_add (jcb, JSC_STATE_PAIR, ss);
if (zhash_insert (ctx->active_jobs, key, (void *)(intptr_t)js) < 0) {
flux_log (h, LOG_ERR, "new_job_cb: inserting a job to hash failed");
goto done;
}
if (invoke_cbs (h, nj, jcb, 0) < 0) {
flux_log (h, LOG_ERR,
"makeup_newjob_event: failed to invoke callbacks");
goto done;
}
done:
Jput (jcb);
free (key);
return;
}
// Recevied a reply to a trigger ("sim.reply")
static void reply_cb (flux_t *h,
flux_msg_handler_t *w,
const flux_msg_t *msg,
void *arg)
{
const char *json_str = NULL;
json_t *request = NULL;
ctx_t *ctx = arg;
sim_state_t *curr_sim_state = ctx->sim_state;
sim_state_t *reply_sim_state;
if (flux_msg_get_json (msg, &json_str) < 0 || json_str == NULL
|| !(request = Jfromstr (json_str))) {
flux_log (h, LOG_ERR, "%s: bad reply message", __FUNCTION__);
Jput (request);
return;
}
// De-serialize and get new info
reply_sim_state = json_to_sim_state (request);
copy_new_state_data (ctx, curr_sim_state, reply_sim_state);
if (handle_next_event (ctx) < 0) {
flux_log (h, LOG_DEBUG, "No events remaining");
if (ctx->exit_on_complete) {
log_msg_exit ("exit_on_complete is set. Exiting now.");
} else {
send_complete_event (h);
}
}
free_simstate (reply_sim_state);
Jput (request);
}
static int update_rdesc (flux_t h, int64_t j, JSON o)
{
int rc = -1;
int64_t nnodes = 0;
int64_t ntasks = 0;
char key1[20] = {'\0'};
char key2[20] = {'\0'};
if (!Jget_int64 (o, JSC_RDESC_NNODES, &nnodes)) return -1;
if (!Jget_int64 (o, JSC_RDESC_NTASKS, &ntasks)) return -1;
if ((nnodes < 0) || (ntasks < 0)) return -1;
snprintf (key1, 20, "lwj.%ld.nnodes", j);
snprintf (key2, 20, "lwj.%ld.ntasks", j);
if (kvs_put_int64 (h, key1, nnodes) < 0)
flux_log (h, LOG_ERR, "update %s: %s", key1, strerror (errno));
else if (kvs_put_int64 (h, key2, ntasks) < 0)
flux_log (h, LOG_ERR, "update %s: %s", key2, strerror (errno));
else if (kvs_commit (h) < 0)
flux_log (h, LOG_ERR, "commit failed");
else {
flux_log (h, LOG_DEBUG, "job (%ld) assigned new resources.", j);
rc = 0;
}
return rc;
}
static int wait_job_complete (flux_t h)
{
int rc = -1;
sig_flux_h = h;
wjctx_t *ctx = getctx (h);
if (signal (SIGINT, sig_handler) == SIG_ERR)
goto done;
if (jsc_notify_status (h, waitjob_cb, (void *)h) != 0) {
flux_log (h, LOG_ERR, "failed to register a waitjob CB");
}
/* once jsc_notify_status is returned, all of JSC events
* will be queued and delivered. It is safe to signal
* readiness.
*/
if (ctx->start)
touch_outfile (ctx->start);
if (complete_job (ctx)) {
if (ctx->complete)
touch_outfile (ctx->complete);
flux_log (ctx->h, LOG_INFO, "wait_job_complete: completion detected");
}
if (flux_reactor_run (flux_get_reactor (h), 0) < 0) {
flux_log (h, LOG_ERR, "error in flux_reactor_run");
goto done;
}
rc = 0;
done:
return rc;
}
/* Accept a connection from new client.
*/
static void listener_cb (flux_t h, flux_fd_watcher_t *w,
int fd, int revents, void *arg)
{
ctx_t *ctx = arg;
if (revents & FLUX_POLLIN) {
client_t *c;
int cfd;
if ((cfd = accept4 (fd, NULL, NULL, SOCK_CLOEXEC)) < 0) {
flux_log (h, LOG_ERR, "accept: %s", strerror (errno));
goto done;
}
if (!(c = client_create (ctx, cfd))) {
close (cfd);
goto done;
}
if (zlist_append (ctx->clients, c) < 0)
oom ();
}
if (revents & ZMQ_POLLERR) {
flux_log (h, LOG_ERR, "poll listen fd: %s", strerror (errno));
}
done:
return;
}
/* Return 'n' sequenced responses.
*/
static int nsrc_request_cb (flux_t h, int typemask, zmsg_t **zmsg, void *arg)
{
JSON o = Jnew ();
int i, count;
if (flux_json_request_decode (*zmsg, &o) < 0) {
if (flux_err_respond (h, errno, zmsg) < 0)
flux_log (h, LOG_ERR, "%s: flux_err_respond: %s", __FUNCTION__,
strerror (errno));
goto done;
}
if (!Jget_int (o, "count", &count)) {
if (flux_err_respond (h, EPROTO, zmsg) < 0)
flux_log (h, LOG_ERR, "%s: flux_err_respond: %s", __FUNCTION__,
strerror (errno));
goto done;
}
for (i = 0; i < count; i++) {
zmsg_t *cpy = zmsg_dup (*zmsg);
if (!cpy)
oom ();
Jadd_int (o, "seq", i);
if (flux_json_respond (h, o, &cpy) < 0)
flux_log (h, LOG_ERR, "%s: flux_json_respond: %s", __FUNCTION__,
strerror (errno));
zmsg_destroy (&cpy);
}
zmsg_destroy (zmsg);
done:
Jput (o);
return 0;
}
static int listener_init (ctx_t *ctx, char *sockpath)
{
struct sockaddr_un addr;
int fd;
fd = socket (AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0);
if (fd < 0) {
flux_log (ctx->h, LOG_ERR, "socket: %s", strerror (errno));
goto done;
}
if (remove (sockpath) < 0 && errno != ENOENT) {
flux_log (ctx->h, LOG_ERR, "remove %s: %s", sockpath, strerror (errno));
goto error_close;
}
memset (&addr, 0, sizeof (struct sockaddr_un));
addr.sun_family = AF_UNIX;
strncpy (addr.sun_path, sockpath, sizeof (addr.sun_path) - 1);
if (bind (fd, (struct sockaddr *)&addr, sizeof (struct sockaddr_un)) < 0) {
flux_log (ctx->h, LOG_ERR, "bind: %s", strerror (errno));
goto error_close;
}
if (listen (fd, LISTEN_BACKLOG) < 0) {
flux_log (ctx->h, LOG_ERR, "listen: %s", strerror (errno));
goto error_close;
}
done:
cleanup_push_string(cleanup_file, sockpath);
return fd;
error_close:
close (fd);
return -1;
}
static int enter_request_cb (flux_t h, int typemask, zmsg_t **zmsg, void *arg)
{
ctx_t *ctx = arg;
barrier_t *b;
json_object *o = NULL;
char *sender = NULL;
const char *name;
int count, nprocs, hopcount;
if (flux_msg_decode (*zmsg, NULL, &o) < 0 || o == NULL
|| !(sender = flux_msg_sender (*zmsg))
|| util_json_object_get_string (o, "name", &name) < 0
|| util_json_object_get_int (o, "count", &count) < 0
|| util_json_object_get_int (o, "nprocs", &nprocs) < 0) {
flux_log (ctx->h, LOG_ERR, "%s: ignoring bad message", __FUNCTION__);
goto done;
}
if (!(b = zhash_lookup (ctx->barriers, name)))
b = barrier_create (ctx, name, nprocs);
/* Distinguish client (tracked) vs downstream barrier plugin (untracked).
* A client, distinguished by hopcount > 0, can only enter barrier once.
*/
if (util_json_object_get_int (o, "hopcount", &hopcount) < 0) {
if (barrier_add_client (b, sender, zmsg) < 0) {
flux_respond_errnum (ctx->h, zmsg, EEXIST);
flux_log (ctx->h, LOG_ERR,
"abort %s due to double entry by client %s",
name, sender);
if (exit_event_send (ctx->h, b->name, ECONNABORTED) < 0)
flux_log (ctx->h, LOG_ERR, "exit_event_send: %s", strerror (errno));
goto done;
}
}
/* If the count has been reached, terminate the barrier;
* o/w set timer to pass count upstream and zero it here.
*/
b->count += count;
if (b->count == b->nprocs) {
if (exit_event_send (ctx->h, b->name, 0) < 0)
flux_log (ctx->h, LOG_ERR, "exit_event_send: %s", strerror (errno));
} else if (!flux_treeroot (ctx->h) && !ctx->timer_armed) {
if (flux_tmouthandler_add (h, barrier_reduction_timeout_msec,
true, timeout_cb, ctx) < 0) {
flux_log (h, LOG_ERR, "flux_tmouthandler_add: %s",strerror (errno));
goto done;
}
ctx->timer_armed = true;
}
done:
if (o)
json_object_put (o);
if (*zmsg)
zmsg_destroy (zmsg);
if (sender)
free (sender);
return 0;
}
static void job_state_cb (flux_t h, flux_msg_handler_t *w,
const flux_msg_t *msg, void *arg)
{
int64_t jobid = -1;
json_object *o = NULL;
const char *topic = NULL;
const char *json_str = NULL;
const char *state = NULL;
int len = 12;
if (flux_msg_get_topic (msg, &topic) < 0)
goto done;
if (flux_event_decode (msg, NULL, &json_str) < 0
|| !(o = Jfromstr (json_str))
|| !Jget_int64 (o, "lwj", &jobid)) {
flux_log (h, LOG_ERR, "%s: bad message", __FUNCTION__);
goto done;
}
if (strncmp (topic, "jsc", 3) == 0)
len = 10;
state = topic + len;
if (strcmp (state, jsc_job_num2state (J_RESERVED)) == 0)
fixup_newjob_event (h, jobid);
if (invoke_cbs (h, jobid, get_update_jcb (h, jobid, state), 0) < 0)
flux_log (h, LOG_ERR, "job_state_cb: failed to invoke callbacks");
if (job_is_finished (state))
delete_jobinfo (h, jobid);
done:
return;
}
int mod_main (flux_t *h, int argc, char **argv)
{
ctx_t *ctx = getctx (h);
uint32_t rank;
flux_msg_handler_t **handlers = NULL;
int rc = -1;
if (flux_get_rank (h, &rank) < 0)
return -1;
if (rank != 0) {
flux_log (h, LOG_ERR, "this module must only run on rank 0");
return -1;
}
flux_log (h, LOG_INFO, "module starting");
if (flux_event_subscribe (h, "sim.start") < 0) {
flux_log (h, LOG_ERR, "subscribing to event: %s", strerror (errno));
return -1;
}
if (flux_msg_handler_addvec (h, htab, ctx, &handlers) < 0) {
flux_log (h, LOG_ERR, "flux_msg_handler_add: %s", strerror (errno));
return -1;
}
send_alive_request (h, module_name);
if (flux_reactor_run (flux_get_reactor (h), 0) < 0) {
flux_log (h, LOG_ERR, "flux_reactor_run: %s", strerror (errno));
goto done_delvec;
}
rc = 0;
done_delvec:
flux_msg_handler_delvec (handlers);
return rc;
}
static int job_status_cb (const char *jcbstr, void *arg, int errnum)
{
int64_t os = 0;
int64_t ns = 0;
int64_t j = 0;
jstatctx_t *ctx = NULL;
flux_t h = (flux_t)arg;
JSON jcb = NULL;
ctx = getctx (h);
if (errnum > 0) {
flux_log (ctx->h, LOG_ERR, "job_status_cb: errnum passed in");
return -1;
}
if (!(jcb = Jfromstr (jcbstr))) {
flux_log (ctx->h, LOG_ERR, "job_status_cb: error parsing JSON string");
return -1;
}
get_jobid (jcb, &j);
get_states (jcb, &os, &ns);
Jput (jcb);
fprintf (ctx->op, "%s->%s\n",
jsc_job_num2state ((job_state_t)os),
jsc_job_num2state ((job_state_t)ns));
fflush (ctx->op);
return 0;
}
static int reg_newjob_hdlr (flux_t h, kvs_set_int64_f func)
{
if (kvs_watch_int64 (h,"lwj.next-id", func, (void *) h) < 0) {
flux_log (h, LOG_ERR, "watch lwj.next-id: %s", strerror (errno));
return -1;
}
flux_log (h, LOG_DEBUG, "registered job creation CB");
return 0;
}
int mod_main (flux_t h, int argc, char **argv)
{
ctx_t *ctx = getctx (h);
char *sockpath = NULL, *dfltpath = NULL;
int rc = -1;
/* Parse args.
*/
if (argc > 0)
sockpath = argv[0];
if (!sockpath)
sockpath = dfltpath = xasprintf ("%s/flux-api", flux_get_tmpdir ());
/* Create listen socket and watcher to handle new connections
*/
if ((ctx->listen_fd = listener_init (ctx, sockpath)) < 0)
goto done;
if (!(ctx->listen_w = flux_fd_watcher_create (ctx->listen_fd,
FLUX_POLLIN | FLUX_POLLERR,
listener_cb, ctx))) {
flux_log (h, LOG_ERR, "flux_fd_watcher_create: %s", strerror (errno));
goto done;
}
flux_fd_watcher_start (h, ctx->listen_w);
/* Create/start event/response message watchers
*/
if (flux_msg_watcher_addvec (h, htab, ctx) < 0) {
flux_log (h, LOG_ERR, "flux_msg_watcher_addvec: %s", strerror (errno));
goto done;
}
/* Start reactor
*/
if (flux_reactor_start (h) < 0) {
flux_log (h, LOG_ERR, "flux_reactor_start: %s", strerror (errno));
goto done;
}
rc = 0;
done:
if (dfltpath)
free (dfltpath);
flux_msg_watcher_delvec (h, htab);
flux_fd_watcher_destroy (ctx->listen_w);
if (ctx->listen_fd >= 0) {
if (close (ctx->listen_fd) < 0)
flux_log (h, LOG_ERR, "close listen_fd: %s", strerror (errno));
}
if (ctx->clients) {
client_t *c;
while ((c = zlist_pop (ctx->clients)))
client_destroy (c);
}
return rc;
}
int sched_plugin_load (struct sched_plugin_loader *sploader, const char *s)
{
char *path = NULL;
char *name = NULL;
char *searchpath = getenv ("FLUX_MODULE_PATH");
void *dso = NULL;
if (sploader->plugin) {
errno = EEXIST;
goto error;
}
if (!searchpath) {
flux_log (sploader->h, LOG_ERR, "FLUX_MODULE_PATH not set");
goto error;
}
if (strchr (s, '/')) {
if (!(name = flux_modname (s))) {
flux_log (sploader->h, LOG_ERR, "%s: %s", s, dlerror ());
errno = ENOENT;
goto error;
}
if (!(path = strdup (s))) {
errno = ENOMEM;
goto error;
}
} else {
if (!(path = flux_modfind (searchpath, s))) {
flux_log (sploader->h, LOG_ERR,
"%s: not found in module search path %s", s, searchpath);
goto error;
}
if (!(name = flux_modname (path)))
goto error;
}
if (!(dso = dlopen (path, RTLD_NOW | RTLD_LOCAL | RTLD_DEEPBIND))) {
flux_log (sploader->h, LOG_ERR, "failed to open sched plugin: %s",
dlerror ());
goto error;
}
flux_log (sploader->h, LOG_DEBUG, "loaded: %s", name);
if (!(sploader->plugin = plugin_create (sploader->h, dso))) {
dlclose (dso);
goto error;
}
sploader->plugin->name = name;
sploader->plugin->path = path;
return 0;
error:
if (path)
free (path);
if (name)
free (name);
return -1;
}
// Looks at the current state and launches the next trigger
static int handle_next_event (ctx_t *ctx)
{
zhash_t *timers;
sim_state_t *sim_state = ctx->sim_state;
int rc = 0;
// get the timer hashtable, make sure its full, and get a list of its keys
timers = sim_state->timers;
if (zhash_size (timers) < 1) {
flux_log (ctx->h, LOG_ERR, "timer hashtable has no elements");
return -1;
}
// Get the next occuring event time/module
double min_event_time = -1;
double *curr_event_time = NULL;
const char *mod_name = NULL, *curr_name = NULL;
for (curr_event_time = zhash_first (timers);
curr_event_time;
curr_event_time = zhash_next (timers)) {
curr_name = zhash_cursor (timers);
if (min_event_time < 0 ||
occurs_before (*curr_event_time, min_event_time) ||
breaks_tie (*curr_event_time, min_event_time, curr_name)) {
min_event_time = *curr_event_time;
mod_name = curr_name;
}
}
if (min_event_time < 0) {
return -1;
}
// advance time then send the trigger to the module with the next event
if (min_event_time > sim_state->sim_time) {
// flux_log (ctx->h, LOG_DEBUG, "Time was advanced from %f to %f while
// triggering the next event for %s",
// sim_state->sim_time, *min_event_time, mod_name);
sim_state->sim_time = min_event_time;
} else {
// flux_log (ctx->h, LOG_DEBUG, "Time was not advanced while triggering
// the next event for %s", mod_name);
}
flux_log (ctx->h,
LOG_DEBUG,
"Triggering %s. Curr sim time: %f",
mod_name,
sim_state->sim_time);
rc = send_trigger (ctx->h, mod_name, sim_state);
return rc;
}
/* Proxy ping.
*/
void xping_request_cb (flux_t h, flux_msg_handler_t *w,
const flux_msg_t *msg, void *arg)
{
ctx_t *ctx = arg;
const char *json_str;
int saved_errno;
int rank, seq = ctx->ping_seq++;
const char *service;
char *hashkey = NULL;
JSON in = Jnew ();
JSON o = NULL;
flux_msg_t *cpy;
if (flux_request_decode (msg, NULL, &json_str) < 0) {
saved_errno = errno;
goto error;
}
if (!(o = Jfromstr (json_str)) || !Jget_int (o, "rank", &rank)
|| !Jget_str (o, "service", &service)) {
saved_errno = errno = EPROTO;
goto error;
}
flux_log (h, LOG_DEBUG, "Rxping rank=%d service=%s", rank, service);
Jadd_int (in, "seq", seq);
flux_log (h, LOG_DEBUG, "Tping seq=%d %d!%s", seq, rank, service);
flux_rpc_t *rpc;
if (!(rpc = flux_rpc (h, service, Jtostr (in), rank,
FLUX_RPC_NORESPONSE))) {
saved_errno = errno;
goto error;
}
flux_rpc_destroy (rpc);
if (!(cpy = flux_msg_copy (msg, true))) {
saved_errno = errno;
goto error;
}
hashkey = xasprintf ("%d", seq);
zhash_update (ctx->ping_requests, hashkey, cpy);
zhash_freefn (ctx->ping_requests, hashkey, (zhash_free_fn *)flux_msg_destroy);
Jput (o);
Jput (in);
if (hashkey)
free (hashkey);
return;
error:
if (flux_respond (h, msg, saved_errno, NULL) < 0)
flux_log_error (h, "%s: flux_respond", __FUNCTION__);
Jput (o);
Jput (in);
}
static int new_job_cb (const char *key, int64_t val, void *arg, int errnum)
{
int64_t nj = 0;
int64_t js = 0;
JSON ss = NULL;
JSON jcb = NULL;
char k[20] = {'\0'};
char path[20] = {'\0'};
flux_t h = (flux_t) arg;
jscctx_t *ctx = getctx (h);
if (ctx->first_time == 1) {
/* watch is invoked immediately and we shouldn't
* rely on that event at all.
*/
ctx->first_time = 0;
return 0;
}
if (chk_errnum (h, errnum) < 0) return 0;
flux_log (h, LOG_DEBUG, "new_job_cb invoked: key(%s), val(%ld)", key, val);
js = J_NULL;
nj = val-1;
snprintf (k, 20, "%ld", nj);
snprintf (path, 20, "lwj.%ld", nj);
if (zhash_insert (ctx->active_jobs, k, (void *)(intptr_t)js) < 0) {
flux_log (h, LOG_ERR, "new_job_cb: inserting a job to hash failed");
goto done;
}
flux_log (h, LOG_DEBUG, "jobstate_hdlr registered");
jcb = Jnew ();
ss = Jnew ();
Jadd_int64 (jcb, JSC_JOBID, nj);
Jadd_int64 (ss, JSC_STATE_PAIR_OSTATE , (int64_t) js);
Jadd_int64 (ss, JSC_STATE_PAIR_NSTATE, (int64_t) js);
json_object_object_add (jcb, JSC_STATE_PAIR, ss);
if (invoke_cbs (h, nj, jcb, errnum) < 0) {
flux_log (h, LOG_ERR, "new_job_cb: failed to invoke callbacks");
}
if (reg_jobstate_hdlr (h, path, job_state_cb) == -1) {
flux_log (h, LOG_ERR, "new_job_cb: reg_jobstate_hdlr: %s",
strerror (errno));
}
done:
/* always return 0 so that reactor won't return */
return 0;
}
static int extract_raw_ntasks (flux_t h, int64_t j, int64_t *ntasks)
{
int rc = 0;
char key[20] = {'\0'};
snprintf (key, 20, "lwj.%ld.ntasks", j);
if (kvs_get_int64 (h, key, ntasks) < 0) {
flux_log (h, LOG_ERR, "extract %s: %s", key, strerror (errno));
rc = -1;
}
else
flux_log (h, LOG_DEBUG, "extract %s: %ld", key, *ntasks);
return rc;
}
static int extract_raw_rdl (flux_t h, int64_t j, char **rdlstr)
{
int rc = 0;
char key[20] = {'\0'};
snprintf (key, 20, "lwj.%ld.rdl", j);
if (kvs_get_string (h, key, rdlstr) < 0) {
flux_log (h, LOG_ERR, "extract %s: %s", key, strerror (errno));
rc = -1;
}
else
flux_log (h, LOG_DEBUG, "rdl under %s extracted", key);
return rc;
}
static int reg_jobstate_hdlr (flux_t h, const char *path, kvs_set_string_f func)
{
int rc = 0;
char key[20] = {'\0'};
snprintf (key, 20, "%s.state", path);
if (kvs_watch_string (h, key, func, (void *)h) < 0) {
flux_log (h, LOG_ERR, "watch %s: %s.", key, strerror (errno));
rc = -1;
} else
flux_log (h, LOG_DEBUG, "registered job %s.state CB", path);
return rc;
}
int mod_main (flux_t h, int argc, char **argv)
{
if (flux_msghandler_add (h, FLUX_MSGTYPE_REQUEST, "job.*",
job_request_cb, NULL) < 0) {
flux_log (h, LOG_ERR, "flux_msghandler_add: %s", strerror (errno));
return -1;
}
if (flux_reactor_start (h) < 0) {
flux_log (h, LOG_ERR, "flux_reactor_start: %s", strerror (errno));
return -1;
}
return 0;
}
int mod_main (flux_t h, int argc, char **argv)
{
flux_flags_set (h, FLUX_O_COPROC);
if (flux_msghandler_addvec (h, htab, htablen, NULL) < 0) {
flux_log (h, LOG_ERR, "flux_msghandler_addvec: %s", strerror (errno));
return -1;
}
if (flux_reactor_start (h) < 0) {
flux_log (h, LOG_ERR, "flux_reactor_start: %s", strerror (errno));
return -1;
}
return 0;
}
/* The req.clog request will not be answered until req.flush is called.
*/
static int stuck_request_cb (flux_t h, int typemask, zmsg_t **zmsg, void *arg)
{
if (flux_json_rpc (h, FLUX_NODEID_ANY, "req.clog", NULL, NULL) < 0) {
flux_log (h, LOG_ERR, "%s: req.clog RPC: %s", __FUNCTION__,
strerror (errno));
return -1;
}
if (flux_err_respond (h, 0, zmsg) < 0) {
flux_log (h, LOG_ERR, "%s: responding: %s", __FUNCTION__,
strerror (errno));
return -1;
}
return 0;
}
int mod_main (flux_t h, int argc, char **argv)
{
ctx_t *ctx = getctx (h);
if (flux_msghandler_addvec (h, htab, htablen, ctx) < 0) {
flux_log (h, LOG_ERR, "flux_msghandler_addvec: %s", strerror (errno));
return -1;
}
if (flux_reactor_start (h) < 0) {
flux_log (h, LOG_ERR, "flux_reactor_start: %s", strerror (errno));
return -1;
}
return 0;
}
static int disconnect (const char *key, void *item, void *arg)
{
barrier_t *b = item;
ctx_t *ctx = b->ctx;
char *sender = arg;
if (zhash_lookup (b->clients, sender)) {
flux_log (ctx->h, LOG_INFO,
"abort %s due to premature disconnect by client %s",
b->name, sender);
if (exit_event_send (ctx->h, b->name, ECONNABORTED) < 0)
flux_log (ctx->h, LOG_ERR, "exit_event_send: %s", strerror (errno));
}
return 0;
}
static int job_state_cb (const char *key, const char *val, void *arg, int errnum)
{
int64_t jobid = -1;
flux_t h = (flux_t) arg;
if (chk_errnum (h, errnum) < 0)
flux_log (h, LOG_ERR, "job_state_cb: key(%s), val(%s)", key, val);
else if (parse_jobid (key, &jobid) != 0)
flux_log (h, LOG_ERR, "job_state_cb: key ill-formed");
else if (invoke_cbs (h, jobid, get_update_jcb (h, jobid, val), errnum) < 0)
flux_log (h, LOG_ERR, "job_state_cb: failed to invoke callbacks");
/* always return 0 so that reactor will not return */
return 0;
}
static client_t * client_create (ctx_t *ctx, int fd)
{
client_t *c;
socklen_t crlen = sizeof (c->ucred);
flux_t h = ctx->h;
c = xzmalloc (sizeof (*c));
c->fd = fd;
if (!(c->uuid = zuuid_new ()))
oom ();
c->ctx = ctx;
if (!(c->disconnect_notify = zhash_new ()))
oom ();
if (!(c->subscriptions = zlist_new ()))
oom ();
if (!(c->outqueue = zlist_new ()))
oom ();
if (getsockopt (fd, SOL_SOCKET, SO_PEERCRED, &c->ucred, &crlen) < 0) {
flux_log (h, LOG_ERR, "getsockopt SO_PEERCRED: %s", strerror (errno));
goto error;
}
assert (crlen == sizeof (c->ucred));
/* Deny connections by uid other than session owner for now.
*/
if (c->ucred.uid != ctx->session_owner) {
flux_log (h, LOG_ERR, "connect by uid=%d pid=%d denied",
c->ucred.uid, (int)c->ucred.pid);
goto error;
}
c->inw = flux_fd_watcher_create (fd, FLUX_POLLIN, client_read_cb, c);
c->outw = flux_fd_watcher_create (fd, FLUX_POLLOUT, client_write_cb, c);
if (!c->inw || !c->outw) {
flux_log (h, LOG_ERR, "flux_fd_watcher_create: %s", strerror (errno));
goto error;
}
flux_fd_watcher_start (h, c->inw);
flux_msg_iobuf_init (&c->inbuf);
flux_msg_iobuf_init (&c->outbuf);
if (set_nonblock (c->fd, true) < 0) {
flux_log (h, LOG_ERR, "set_nonblock: %s", strerror (errno));
goto error;
}
return (c);
error:
client_destroy (c);
return NULL;
}
/* Received response message from broker.
* Look up the sender uuid in clients hash and deliver.
* Responses for disconnected clients are silently discarded.
*/
static void response_cb (flux_t h, flux_msg_watcher_t *w,
const flux_msg_t *msg, void *arg)
{
ctx_t *ctx = arg;
char *uuid = NULL;
client_t *c;
flux_msg_t *cpy = flux_msg_copy (msg, true);
if (!cpy)
oom ();
if (flux_msg_pop_route (cpy, &uuid) < 0)
goto done;
if (flux_msg_clear_route (cpy) < 0)
goto done;
c = zlist_first (ctx->clients);
while (c) {
if (!strcmp (uuid, zuuid_str (c->uuid))) {
if (client_send_nocopy (c, &cpy) < 0) { /* FIXME handle errors */
flux_log (h, LOG_ERR, "%s: client_send %s: %s",
__FUNCTION__, zuuid_str (c->uuid), strerror (errno));
errno = 0;
}
break;
}
c = zlist_next (ctx->clients);
}
if (uuid)
free (uuid);
done:
flux_msg_destroy (cpy);
}
