From 94e863de168a201a4b72aa08e70dd83836766ae0 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Thu, 13 Jul 2006 19:14:45 +0000
Subject: [PATCH] svn merge -r8549:8565
https://eris.llnl.gov/svn/slurm/branches/slurm-1.1
---
src/common/forward.c | 5 ++
src/common/hostlist.c | 75 +++++++++++++++-----
src/common/slurm_protocol_api.c | 4 +-
src/common/slurm_protocol_defs.h | 1 +
src/common/slurm_protocol_pack.c | 3 +
src/sbcast/agent.c | 5 +-
src/sinfo/opts.c | 14 +++-
src/slurmctld/agent.c | 114 ++++++++++++++++++++++++-------
8 files changed, 176 insertions(+), 45 deletions(-)
diff --git a/src/common/forward.c b/src/common/forward.c
index e922ba27640..817d1a59db2 100644
--- a/src/common/forward.c
+++ b/src/common/forward.c
@@ -497,6 +497,8 @@ extern int forward_msg_to_next(forward_msg_t *fwd_msg, int err)
if(fwd_msg->ret_list) {
ret_data_info = xmalloc(sizeof(ret_data_info_t));
ret_data_info->node_name = xstrdup(fwd_msg->node_name);
+ memcpy(&ret_data_info->addr, &fwd_msg->addr,
+ sizeof(slurm_addr));
ret_data_info->nodeid = fwd_msg->header.srun_node_id;
itr = list_iterator_create(fwd_msg->ret_list);
while((type = (ret_types_t *) list_next(itr)) != NULL) {
@@ -747,6 +749,8 @@ extern int no_resp_forwards(forward_t *forward, List *ret_list, int err)
&forward->name[i * MAX_SLURM_NAME],
MAX_SLURM_NAME);
ret_data_info->node_name = xstrdup(name);
+ memcpy(&ret_data_info->addr, &forward->addr[i],
+ sizeof(slurm_addr));
ret_data_info->nodeid = forward->node_id[i];
}
no_forward:
@@ -769,6 +773,7 @@ void destroy_forward(forward_t *forward)
xfree(forward->name);
xfree(forward->node_id);
forward->cnt = 0;
+ forward->init = 0;
}
}
diff --git a/src/common/hostlist.c b/src/common/hostlist.c
index c02420ff26c..8010b1a3d9a 100644
--- a/src/common/hostlist.c
+++ b/src/common/hostlist.c
@@ -519,7 +519,10 @@ static int _width_equiv(unsigned long n, int *wn, unsigned long m, int *wm)
*/
static size_t host_prefix_end(const char *hostname)
{
- size_t idx = strlen(hostname) - 1;
+ size_t idx;
+ if (!hostname)
+ return -1;
+ idx = strlen(hostname) - 1;
while (idx >= 0 && isdigit((char) hostname[idx]))
idx--;
@@ -598,6 +601,8 @@ static void hostname_destroy(hostname_t hn)
*/
static int hostname_suffix_is_valid(hostname_t hn)
{
+ if (!hn)
+ return false;
return hn->suffix != NULL;
}
@@ -605,6 +610,8 @@ static int hostname_suffix_is_valid(hostname_t hn)
*/
static int hostname_suffix_width(hostname_t hn)
{
+ if (!hn)
+ return -1;
assert(hn->suffix != NULL);
return (int) strlen(hn->suffix);
}
@@ -768,12 +775,12 @@ static int hostrange_cmp(hostrange_t h1, hostrange_t h2)
/* compare the prefixes of two hostrange objects.
* returns:
- * < 0 if h1 prefix is less than h2 OR h1 == NULL.
+ * < 0 if h1 prefix is less than h2 OR h2 == NULL.
*
* 0 if h1's prefix and h2's prefix match,
* UNLESS, either h1 or h2 (NOT both) do not have a valid suffix.
*
- * > 0 if h1's prefix is greater than h2's OR h2 == NULL. */
+ * > 0 if h1's prefix is greater than h2's OR h1 == NULL. */
static int hostrange_prefix_cmp(hostrange_t h1, hostrange_t h2)
{
int retval;
@@ -982,6 +989,8 @@ hostrange_to_string(hostrange_t hr, size_t n, char *buf, char *separator)
if (n == 0)
return 0;
+
+ assert(hr != NULL);
if (hr->singlehost)
return snprintf(buf, n, "%s", hr->prefix);
@@ -1017,6 +1026,7 @@ static size_t hostrange_numstr(hostrange_t hr, size_t n, char *buf)
int len = 0;
assert(buf != NULL);
+ assert(hr != NULL);
if (hr->singlehost || n == 0)
return 0;
@@ -1245,8 +1255,11 @@ hostlist_t _hostlist_create(const char *hostlist, char *sep, char *r_op)
hostlist_t new = hostlist_new();
- orig = str = strdup(hostlist);
+ if (hostlist == NULL)
+ return new;
+ orig = str = strdup(hostlist);
+
/* return an empty list if an empty string was passed in */
if (str == NULL || strlen(str) == 0)
goto done;
@@ -1356,7 +1369,8 @@ hostlist_t _hostlist_create(const char *hostlist, char *sep, char *r_op)
}
done:
- free(orig);
+ if(orig)
+ free(orig);
return new;
}
@@ -1412,7 +1426,7 @@ static int _parse_single_range(const char *str, struct _range *range)
free(orig);
range->width = strlen(str);
return 1;
-
+
error:
errno = EINVAL;
_error(__FILE__, __LINE__, "Invalid range: `%s'", orig);
@@ -1499,6 +1513,7 @@ _push_range_list(hostlist_t hl, char *pfx, struct _range *rng,
int n)
{
int i;
+
for (i = 0; i < n; i++) {
hostlist_push_hr(hl, pfx, rng->lo, rng->hi, rng->width);
rng++;
@@ -1571,7 +1586,7 @@ hostlist_t hostlist_copy(const hostlist_t hl)
int i;
hostlist_t new;
- if (hl == NULL)
+ if (!hl)
return NULL;
LOCK_HOSTLIST(hl);
@@ -1595,7 +1610,7 @@ hostlist_t hostlist_copy(const hostlist_t hl)
void hostlist_destroy(hostlist_t hl)
{
int i;
- if (hl == NULL)
+ if (!hl)
return;
LOCK_HOSTLIST(hl);
while (hl->ilist) {
@@ -1617,7 +1632,7 @@ int hostlist_push(hostlist_t hl, const char *hosts)
{
hostlist_t new;
int retval;
- if (hosts == NULL)
+ if (!hosts || !hl)
return 0;
new = hostlist_create(hosts);
if (!new)
@@ -1635,7 +1650,7 @@ int hostlist_push_host(hostlist_t hl, const char *str)
hostrange_t hr;
hostname_t hn;
- if (str == NULL)
+ if (!str || !hl)
return 0;
hn = hostname_create(str);
@@ -1658,7 +1673,7 @@ int hostlist_push_list(hostlist_t h1, hostlist_t h2)
{
int i, n = 0;
- if (h2 == NULL)
+ if (!h2 || !h1)
return 0;
LOCK_HOSTLIST(h2);
@@ -1675,7 +1690,11 @@ int hostlist_push_list(hostlist_t h1, hostlist_t h2)
char *hostlist_pop(hostlist_t hl)
{
char *host = NULL;
-
+ if(!hl) {
+ error("hostlist_pop: no hoslist given");
+ return NULL;
+ }
+
LOCK_HOSTLIST(hl);
if (hl->nhosts > 0) {
hostrange_t hr = hl->hr[hl->nranges - 1];
@@ -1696,6 +1715,10 @@ static void
hostlist_shift_iterators(hostlist_t hl, int idx, int depth, int n)
{
hostlist_iterator_t i;
+ if(!hl) {
+ error("hostlist_shift_iterators: no hoslist given");
+ return;
+ }
for (i = hl->ilist; i; i = i->next) {
if (n == 0) {
if (i->idx == idx && i->depth >= depth)
@@ -1715,6 +1738,10 @@ char *hostlist_shift(hostlist_t hl)
{
char *host = NULL;
+ if(!hl){
+ error("hostlist_shift: no hoslist given");
+ return NULL;
+ }
LOCK_HOSTLIST(hl);
if (hl->nhosts > 0) {
@@ -1743,6 +1770,8 @@ char *hostlist_pop_range(hostlist_t hl)
hostlist_t hltmp;
hostrange_t tail;
+ if(!hl)
+ return NULL;
LOCK_HOSTLIST(hl);
if (hl->nranges < 1 || !(hltmp = hostlist_new())) {
UNLOCK_HOSTLIST(hl);
@@ -1774,7 +1803,7 @@ char *hostlist_shift_range(hostlist_t hl)
int i;
char buf[1024];
hostlist_t hltmp = hostlist_new();
- if (!hltmp)
+ if (!hltmp || !hl)
return NULL;
LOCK_HOSTLIST(hl);
@@ -1816,7 +1845,9 @@ int hostlist_delete(hostlist_t hl, const char *hosts)
int n = 0;
char *hostname = NULL;
hostlist_t hltmp;
-
+ if(!hl)
+ return -1;
+
if (!(hltmp = hostlist_create(hosts)))
seterrno_ret(EINVAL, 0);
@@ -1833,7 +1864,12 @@ int hostlist_delete(hostlist_t hl, const char *hosts)
/* XXX watch out! poor implementation follows! (fix it at some point) */
int hostlist_delete_host(hostlist_t hl, const char *hostname)
{
- int n = hostlist_find(hl, hostname);
+ int n;
+
+ if(!hl)
+ return -1;
+ n = hostlist_find(hl, hostname);
+
if (n >= 0)
hostlist_delete_nth(hl, n);
return n >= 0 ? 1 : 0;
@@ -1857,6 +1893,8 @@ char * hostlist_nth(hostlist_t hl, int n)
char *host = NULL;
int i, count;
+ if(!hl)
+ return NULL;
LOCK_HOSTLIST(hl);
count = 0;
for (i = 0; i < hl->nranges; i++) {
@@ -1879,6 +1917,8 @@ int hostlist_delete_nth(hostlist_t hl, int n)
{
int i, count;
+ if(!hl)
+ return -1;
LOCK_HOSTLIST(hl);
assert(n >= 0 && n <= hl->nhosts);
@@ -1915,6 +1955,9 @@ int hostlist_delete_nth(hostlist_t hl, int n)
int hostlist_count(hostlist_t hl)
{
int retval;
+ if(!hl)
+ return -1;
+
LOCK_HOSTLIST(hl);
retval = hl->nhosts;
UNLOCK_HOSTLIST(hl);
@@ -1926,7 +1969,7 @@ int hostlist_find(hostlist_t hl, const char *hostname)
int i, count, ret = -1;
hostname_t hn;
- if (!hostname)
+ if (!hostname || !hl)
return -1;
hn = hostname_create(hostname);
diff --git a/src/common/slurm_protocol_api.c b/src/common/slurm_protocol_api.c
index 20c42298edf..7baabf5edae 100644
--- a/src/common/slurm_protocol_api.c
+++ b/src/common/slurm_protocol_api.c
@@ -1578,7 +1578,7 @@ static List _send_recv_rc_msg(slurm_fd fd, slurm_msg_t *req, int timeout)
}
ret_data_info = xmalloc(sizeof(ret_data_info_t));
- ret_data_info->node_name = xstrdup("localhost");
+ ret_data_info->node_name = NULL;
ret_data_info->data = NULL;
debug3("got reply for %s rc %d %d",
ret_data_info->node_name,
@@ -1674,7 +1674,7 @@ failed:
}
ret_data_info = xmalloc(sizeof(ret_data_info_t));
- ret_data_info->node_name = xstrdup("localhost");
+ ret_data_info->node_name = NULL;
ret_data_info->data = NULL;
itr = list_iterator_create(ret_list);
while((ret_type = list_next(itr)) != NULL) {
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index cf10651182a..47b686dc63a 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -252,6 +252,7 @@ typedef struct slurm_msg {
typedef struct ret_data_info {
char *node_name;
+ slurm_addr addr;
uint32_t nodeid;
void *data;
} ret_data_info_t;
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index d24e89dbf1a..7e2e82eec37 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -3121,6 +3121,7 @@ _pack_ret_list(List ret_list,
itr_data = list_iterator_create(ret_type->ret_data_list);
while((ret_data_info = list_next(itr_data)) != NULL) {
packstr(ret_data_info->node_name, buffer);
+ slurm_pack_slurm_addr(&ret_data_info->addr, buffer);
pack32((uint32_t)ret_data_info->nodeid, buffer);
msg.data = ret_data_info->data;
pack_msg(&msg, buffer);
@@ -3156,6 +3157,8 @@ _unpack_ret_list(List *ret_list,
ret_data_info = xmalloc(sizeof(ret_data_info_t));
safe_unpackstr_xmalloc(&ret_data_info->node_name,
&uint16_tmp, buffer);
+ slurm_unpack_slurm_addr_no_alloc(&ret_data_info->addr,
+ buffer);
safe_unpack32((uint32_t *)&ret_data_info->nodeid,
buffer);
if (unpack_msg(&msg, buffer) != SLURM_SUCCESS)
diff --git a/src/sbcast/agent.c b/src/sbcast/agent.c
index d2a52fde1e0..22d292188c5 100644
--- a/src/sbcast/agent.c
+++ b/src/sbcast/agent.c
@@ -88,11 +88,10 @@ static void *_agent_thread(void *args)
while ((ret_data_info = list_next(data_itr)) != NULL) {
if (ret_type->msg_rc == SLURM_SUCCESS)
continue;
- if (!strcmp(ret_data_info->node_name,
- "localhost")) {
- xfree(ret_data_info->node_name);
+ if (!ret_data_info->node_name) {
ret_data_info->node_name =
xstrdup(thread_ptr->node_name);
+ ret_data_info->addr = msg->address;
}
error("REQUEST_FILE_BCAST(%s): %s",
ret_data_info->node_name,
diff --git a/src/sinfo/opts.c b/src/sinfo/opts.c
index d6d0d6848ff..a7002ee7f9b 100644
--- a/src/sinfo/opts.c
+++ b/src/sinfo/opts.c
@@ -73,6 +73,7 @@ extern void parse_command_line(int argc, char *argv[])
char *env_val = NULL;
int opt_char;
int option_index;
+ hostlist_t host_list;
static struct option long_options[] = {
{"all", no_argument, 0, 'a'},
{"bg", no_argument, 0, 'b'},
@@ -136,7 +137,8 @@ extern void parse_command_line(int argc, char *argv[])
case (int) 'i':
params.iterate= atoi(optarg);
if (params.iterate <= 0) {
- error ("Error: --iterate=%s");
+ error ("Error: invalid entry for "
+ "--iterate=%s", optarg);
exit(1);
}
break;
@@ -145,6 +147,16 @@ extern void parse_command_line(int argc, char *argv[])
break;
case (int) 'n':
params.nodes= xstrdup(optarg);
+ /*
+ * confirm valid nodelist entry
+ */
+ host_list = hostlist_create(params.nodes);
+ if (!host_list) {
+ error("'%s' invalid entry for --nodes",
+ optarg);
+ exit(1);
+ }
+ hostlist_destroy(host_list);
break;
case (int) 'N':
params.node_flag = true;
diff --git a/src/slurmctld/agent.c b/src/slurmctld/agent.c
index 07d7861a9c8..2d47e1ddfde 100644
--- a/src/slurmctld/agent.c
+++ b/src/slurmctld/agent.c
@@ -161,6 +161,8 @@ static void _notify_slurmctld_nodes(agent_info_t *agent_ptr,
int no_resp_cnt, int retry_cnt);
static void _purge_agent_args(agent_arg_t *agent_arg_ptr);
static void _queue_agent_retry(agent_info_t * agent_info_ptr, int count);
+static int _setup_requeue(agent_arg_t *agent_arg_ptr, thd_t *thread_ptr,
+ int count, int *spot);
static void _slurmctld_free_job_launch_msg(batch_job_launch_msg_t * msg);
static void _spawn_retry_agent(agent_arg_t * agent_arg_ptr);
static void *_thread_per_group_rpc(void *args);
@@ -455,7 +457,7 @@ static void _update_wdog_state(thd_t *thread_ptr,
static void *_wdog(void *args)
{
bool srun_agent = false;
- int i;
+ int i, j, count;
agent_info_t *agent_ptr = (agent_info_t *) args;
thd_t *thread_ptr = agent_ptr->thread_struct;
unsigned long usec = 1250000;
@@ -484,6 +486,7 @@ static void *_wdog(void *args)
slurm_mutex_lock(&agent_ptr->thread_mutex);
for (i = 0; i < agent_ptr->thread_count; i++) {
+ //info("thread name %s",thread_ptr[i].node_name);
if(!thread_ptr[i].ret_list) {
_update_wdog_state(&thread_ptr[i],
&thread_ptr[i].state,
@@ -492,10 +495,14 @@ static void *_wdog(void *args)
itr = list_iterator_create(
thread_ptr[i].ret_list);
while((ret_type = list_next(itr)) != NULL) {
- _update_wdog_state(
- &thread_ptr[i],
- (state_t *)&ret_type->msg_rc,
- &thd_comp);
+ count = list_count(ret_type->
+ ret_data_list);
+ for(j=0; j<count; j++) {
+ _update_wdog_state(
+ &thread_ptr[i],
+ &ret_type->msg_rc,
+ &thd_comp);
+ }
}
list_iterator_destroy(itr);
}
@@ -518,7 +525,7 @@ static void *_wdog(void *args)
if (thread_ptr[i].ret_list)
list_destroy(thread_ptr[i].ret_list);
}
-
+
if (thd_comp.max_delay)
debug2("agent maximum delay %d seconds", thd_comp.max_delay);
@@ -817,7 +824,7 @@ static void *_thread_per_group_rpc(void *args)
msg.srun_node_id = 0;
msg.forward_struct_init = 0;
- //info("forwarding to %d",msg.forward.cnt);
+ //info("%s forwarding to %d",thread_ptr->node_name, msg.forward.cnt);
thread_ptr->end_time = thread_ptr->start_time + COMMAND_TIMEOUT;
if (task_ptr->get_reply) {
send_rc_again:
@@ -844,6 +851,9 @@ static void *_thread_per_group_rpc(void *args)
strncpy(thread_ptr->node_name,
fwd_msg.node_name,
MAX_SLURM_NAME);
+ memcpy(&thread_ptr->slurm_addr,
+ &fwd_msg.addr,
+ sizeof(slurm_addr));
goto send_rc_again;
}
}
@@ -854,7 +864,8 @@ static void *_thread_per_group_rpc(void *args)
tmp_ret_list = list_create(destroy_ret_types);
fwd_msg.header.srun_node_id = msg.srun_node_id;
- fwd_msg.header.forward = msg.forward;
+ forward_init(&fwd_msg.header.forward, &msg.forward);
+ //fwd_msg.header.forward = msg.forward;
fwd_msg.ret_list = tmp_ret_list;
strncpy(fwd_msg.node_name,
thread_ptr->node_name,
@@ -862,11 +873,15 @@ static void *_thread_per_group_rpc(void *args)
fwd_msg.forward_mutex = NULL;
if(forward_msg_to_next(&fwd_msg, errno)) {
msg.address = fwd_msg.addr;
- msg.forward = fwd_msg.header.forward;
+ forward_init(&msg.forward,
+ &fwd_msg.header.forward);
msg.srun_node_id = fwd_msg.header.srun_node_id;
strncpy(thread_ptr->node_name,
fwd_msg.node_name,
MAX_SLURM_NAME);
+ memcpy(&thread_ptr->slurm_addr,
+ &fwd_msg.addr,
+ sizeof(slurm_addr));
goto send_node_again;
}
}
@@ -894,15 +909,17 @@ static void *_thread_per_group_rpc(void *args)
data_itr = list_iterator_create(ret_type->ret_data_list);
while((ret_data_info = list_next(data_itr)) != NULL) {
rc = ret_type->msg_rc;
- if(!found
- && !strcmp(ret_data_info->node_name,"localhost")) {
- //info("got localhost");
- xfree(ret_data_info->node_name);
+ if(!found && !ret_data_info->node_name) {
ret_data_info->node_name =
xstrdup(thread_ptr->node_name);
+ memcpy(&ret_data_info->addr,
+ &thread_ptr->slurm_addr,
+ sizeof(slurm_addr));
+ /* info("got localhost changing to %s", */
+/* ret_data_info->node_name); */
found = 1;
}
-/* info("response for %s rc = %d", */
+ /* info("response for %s rc = %d", */
/* ret_data_info->node_name, */
/* ret_type->msg_rc); */
if(rc == SLURM_ERROR) {
@@ -1006,7 +1023,7 @@ static void *_thread_per_group_rpc(void *args)
ret_data_info->node_name);
}
list_iterator_destroy(data_itr);
- if (srun_agent)
+ if(srun_agent)
thread_state = DSH_FAILED;
else if(ret_type->type == REQUEST_PING)
/* check if a forward failed */
@@ -1024,6 +1041,8 @@ static void *_thread_per_group_rpc(void *args)
cleanup:
xfree(args);
+
+ /* handled at end of thread just incase resend is needed */
destroy_forward(&msg.forward);
slurm_mutex_lock(thread_mutex_ptr);
thread_ptr->ret_list = ret_list;
@@ -1046,6 +1065,45 @@ static void _alarm_handler(int dummy)
xsignal(SIGALRM, _alarm_handler);
}
+static int _setup_requeue(agent_arg_t *agent_arg_ptr, thd_t *thread_ptr,
+ int count, int *spot)
+{
+ ListIterator itr;
+ ListIterator data_itr;
+ ret_data_info_t *ret_data_info = NULL;
+ ret_types_t *ret_type = NULL;
+
+ itr = list_iterator_create(thread_ptr->ret_list);
+ while((ret_type = list_next(itr)) != NULL) {
+ debug2("got return type of %d", ret_type->msg_rc);
+ if (ret_type->msg_rc != DSH_NO_RESP)
+ continue;
+
+ data_itr = list_iterator_create(ret_type->ret_data_list);
+ while((ret_data_info = list_next(data_itr)) != NULL) {
+ debug("got the name %s to resend out of %d",
+ ret_data_info->node_name, count);
+
+ if(agent_arg_ptr) {
+ memcpy(&agent_arg_ptr->slurm_addr[*spot],
+ &ret_data_info->addr,
+ sizeof(slurm_addr));
+
+ strncpy(&agent_arg_ptr->
+ node_names[(*spot) * MAX_SLURM_NAME],
+ ret_data_info->node_name,
+ MAX_SLURM_NAME);
+
+ if ((++(*spot)) == count)
+ return 1;
+ }
+ }
+ list_iterator_destroy(data_itr);
+ }
+ list_iterator_destroy(itr);
+ return 0;
+}
+
/*
* _queue_agent_retry - Queue any failed RPCs for later replay
* IN agent_info_ptr - pointer to info on completed agent requests
@@ -1057,7 +1115,7 @@ static void _queue_agent_retry(agent_info_t * agent_info_ptr, int count)
queued_request_t *queued_req_ptr = NULL;
thd_t *thread_ptr = agent_info_ptr->thread_struct;
int i, j;
-
+
if (count == 0)
return;
@@ -1074,13 +1132,23 @@ static void _queue_agent_retry(agent_info_t * agent_info_ptr, int count)
j = 0;
for (i = 0; i < agent_info_ptr->thread_count; i++) {
- if (thread_ptr[i].state != DSH_NO_RESP)
- continue;
- agent_arg_ptr->slurm_addr[j] = thread_ptr[i].slurm_addr;
- strncpy(&agent_arg_ptr->node_names[j * MAX_SLURM_NAME],
- thread_ptr[i].node_name, MAX_SLURM_NAME);
- if ((++j) == count)
- break;
+ if(!thread_ptr[i].ret_list) {
+ if (thread_ptr[i].state != DSH_NO_RESP)
+ continue;
+ debug("got the name %s to resend",
+ thread_ptr[i].node_name);
+
+ agent_arg_ptr->slurm_addr[j] =
+ thread_ptr[i].slurm_addr;
+ strncpy(&agent_arg_ptr->node_names[j * MAX_SLURM_NAME],
+ thread_ptr[i].node_name, MAX_SLURM_NAME);
+ if ((++j) == count)
+ break;
+ } else {
+ if(_setup_requeue(agent_arg_ptr, &thread_ptr[i],
+ count, &j))
+ break;
+ }
}
if (count != j) {
error("agent: Retry count (%d) != actual count (%d)",
--
GitLab