diff --git a/src/common/checkpoint.c b/src/common/checkpoint.c
index 347c95f24cb6f1ced076880ec82112726ba5eb92..eee6ba454f6a70b794b278057edfa312c44c3966 100644
--- a/src/common/checkpoint.c
+++ b/src/common/checkpoint.c
@@ -72,8 +72,10 @@ typedef struct slurm_checkpoint_ops {
int (*ckpt_alloc_jobinfo) (check_jobinfo_t *jobinfo);
int (*ckpt_free_jobinfo) (check_jobinfo_t jobinfo);
- int (*ckpt_pack_jobinfo) (check_jobinfo_t jobinfo, Buf buffer);
- int (*ckpt_unpack_jobinfo) (check_jobinfo_t jobinfo, Buf buffer);
+ int (*ckpt_pack_jobinfo) (check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version);
+ int (*ckpt_unpack_jobinfo) (check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version);
int (*ckpt_stepd_prefork) (void *slurmd_job);
int (*ckpt_signal_tasks) (void *slurmd_job, char *image_dir);
int (*ckpt_restart_task) (void *slurmd_job, char *image_dir, int gtid);
@@ -368,14 +370,15 @@ extern int checkpoint_free_jobinfo(check_jobinfo_t jobinfo)
}
/* un/pack a job step's checkpoint context */
-extern int checkpoint_pack_jobinfo (check_jobinfo_t jobinfo, Buf buffer)
+extern int checkpoint_pack_jobinfo (check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
int retval = SLURM_SUCCESS;
slurm_mutex_lock( &context_lock );
if ( g_context )
retval = (*(g_context->ops.ckpt_pack_jobinfo))(
- jobinfo, buffer);
+ jobinfo, buffer, protocol_version);
else {
error ("slurm_checkpoint plugin context not initialized");
retval = ENOENT;
@@ -384,14 +387,15 @@ extern int checkpoint_pack_jobinfo (check_jobinfo_t jobinfo, Buf buffer)
return retval;
}
-extern int checkpoint_unpack_jobinfo (check_jobinfo_t jobinfo, Buf buffer)
+extern int checkpoint_unpack_jobinfo (check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
int retval = SLURM_SUCCESS;
slurm_mutex_lock( &context_lock );
if ( g_context )
retval = (*(g_context->ops.ckpt_unpack_jobinfo))(
- jobinfo, buffer);
+ jobinfo, buffer, protocol_version);
else {
error ("slurm_checkpoint plugin context not initialized");
retval = ENOENT;
diff --git a/src/common/checkpoint.h b/src/common/checkpoint.h
index d65c22f335fe537d41474613d5d65744f9ff9abf..b7bfc90fe4332dbf0ae6c0c995ded82687e270ea 100644
--- a/src/common/checkpoint.h
+++ b/src/common/checkpoint.h
@@ -79,11 +79,12 @@ extern int checkpoint_op(uint32_t job_id, uint32_t step_id,
uint32_t *error_code, char **error_msg);
/* note checkpoint completion */
-extern int checkpoint_comp(void * step_ptr, time_t event_time, uint32_t error_code,
- char *error_msg);
+extern int checkpoint_comp(void * step_ptr, time_t event_time,
+ uint32_t error_code, char *error_msg);
extern int checkpoint_task_comp(void * step_ptr, uint32_t task_id,
- time_t event_time, uint32_t error_code, char *error_msg);
+ time_t event_time, uint32_t error_code,
+ char *error_msg);
/* gather checkpoint error info */
extern int checkpoint_error(void * step_ptr,
@@ -96,8 +97,10 @@ extern int checkpoint_alloc_jobinfo(check_jobinfo_t *jobinfo);
extern int checkpoint_free_jobinfo(check_jobinfo_t jobinfo);
/* un/pack a job step's checkpoint context */
-extern int checkpoint_pack_jobinfo (check_jobinfo_t jobinfo, Buf buffer);
-extern int checkpoint_unpack_jobinfo (check_jobinfo_t jobinfo, Buf buffer);
+extern int checkpoint_pack_jobinfo (check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version);
+extern int checkpoint_unpack_jobinfo (check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version);
/* create the necessary threads before forking the tasks */
extern int checkpoint_stepd_prefork (void *slurmd_job);
diff --git a/src/plugins/checkpoint/aix/checkpoint_aix.c b/src/plugins/checkpoint/aix/checkpoint_aix.c
index 04fcdbaff9e7dcc7ca5f886528399cd5acd5b863..41baffc84d0795e6fee02f72e7b17d466a9e61ff 100644
--- a/src/plugins/checkpoint/aix/checkpoint_aix.c
+++ b/src/plugins/checkpoint/aix/checkpoint_aix.c
@@ -311,37 +311,44 @@ extern int slurm_ckpt_free_job(check_jobinfo_t jobinfo)
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
struct check_job_info *check_ptr =
(struct check_job_info *)jobinfo;
- pack16(check_ptr->disabled, buffer);
- pack16(check_ptr->node_cnt, buffer);
- pack16(check_ptr->reply_cnt, buffer);
- pack16(check_ptr->wait_time, buffer);
+ if(protocol_version >= SLURM_2_1_PROTOCOL_VERSION) {
+ pack16(check_ptr->disabled, buffer);
+ pack16(check_ptr->node_cnt, buffer);
+ pack16(check_ptr->reply_cnt, buffer);
+ pack16(check_ptr->wait_time, buffer);
- pack32(check_ptr->error_code, buffer);
- packstr(check_ptr->error_msg, buffer);
- pack_time(check_ptr->time_stamp, buffer);
+ pack32(check_ptr->error_code, buffer);
+ packstr(check_ptr->error_msg, buffer);
+ pack_time(check_ptr->time_stamp, buffer);
+ }
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
uint32_t uint32_tmp;
struct check_job_info *check_ptr =
(struct check_job_info *)jobinfo;
- safe_unpack16(&check_ptr->disabled, buffer);
- safe_unpack16(&check_ptr->node_cnt, buffer);
- safe_unpack16(&check_ptr->reply_cnt, buffer);
- safe_unpack16(&check_ptr->wait_time, buffer);
+ if(protocol_version >= SLURM_2_1_PROTOCOL_VERSION) {
+ safe_unpack16(&check_ptr->disabled, buffer);
+ safe_unpack16(&check_ptr->node_cnt, buffer);
+ safe_unpack16(&check_ptr->reply_cnt, buffer);
+ safe_unpack16(&check_ptr->wait_time, buffer);
- safe_unpack32(&check_ptr->error_code, buffer);
- safe_unpackstr_xmalloc(&check_ptr->error_msg, &uint32_tmp, buffer);
- safe_unpack_time(&check_ptr->time_stamp, buffer);
+ safe_unpack32(&check_ptr->error_code, buffer);
+ safe_unpackstr_xmalloc(&check_ptr->error_msg,
+ &uint32_tmp, buffer);
+ safe_unpack_time(&check_ptr->time_stamp, buffer);
+ }
return SLURM_SUCCESS;
diff --git a/src/plugins/checkpoint/blcr/checkpoint_blcr.c b/src/plugins/checkpoint/blcr/checkpoint_blcr.c
index 05fe622bd73c6367d63cbe4775db3e439051131d..86a4b677c5ea2047b8f4b10623fe6fbed0fc26da 100644
--- a/src/plugins/checkpoint/blcr/checkpoint_blcr.c
+++ b/src/plugins/checkpoint/blcr/checkpoint_blcr.c
@@ -321,29 +321,36 @@ extern int slurm_ckpt_free_job(check_jobinfo_t jobinfo)
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
struct check_job_info *check_ptr =
(struct check_job_info *)jobinfo;
- pack16(check_ptr->disabled, buffer);
- pack_time(check_ptr->time_stamp, buffer);
- pack32(check_ptr->error_code, buffer);
- packstr(check_ptr->error_msg, buffer);
+ if(protocol_version >= SLURM_2_1_PROTOCOL_VERSION) {
+ pack16(check_ptr->disabled, buffer);
+ pack_time(check_ptr->time_stamp, buffer);
+ pack32(check_ptr->error_code, buffer);
+ packstr(check_ptr->error_msg, buffer);
+ }
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
uint32_t uint32_tmp;
struct check_job_info *check_ptr =
(struct check_job_info *)jobinfo;
- safe_unpack16(&check_ptr->disabled, buffer);
- safe_unpack_time(&check_ptr->time_stamp, buffer);
- safe_unpack32(&check_ptr->error_code, buffer);
- safe_unpackstr_xmalloc(&check_ptr->error_msg, &uint32_tmp, buffer);
+ if(protocol_version >= SLURM_2_1_PROTOCOL_VERSION) {
+ safe_unpack16(&check_ptr->disabled, buffer);
+ safe_unpack_time(&check_ptr->time_stamp, buffer);
+ safe_unpack32(&check_ptr->error_code, buffer);
+ safe_unpackstr_xmalloc(&check_ptr->error_msg,
+ &uint32_tmp, buffer);
+ }
return SLURM_SUCCESS;
diff --git a/src/plugins/checkpoint/none/checkpoint_none.c b/src/plugins/checkpoint/none/checkpoint_none.c
index 4e4259c51bd31218476900241ec0bc3312196321..253c62a71c21b9ebec647c349d304ae8c9014c65 100644
--- a/src/plugins/checkpoint/none/checkpoint_none.c
+++ b/src/plugins/checkpoint/none/checkpoint_none.c
@@ -128,12 +128,14 @@ extern int slurm_ckpt_free_job(check_jobinfo_t jobinfo)
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
return SLURM_SUCCESS;
}
diff --git a/src/plugins/checkpoint/ompi/checkpoint_ompi.c b/src/plugins/checkpoint/ompi/checkpoint_ompi.c
index a69ce989aa9ddb67995a0d68b7e2e58839983c8a..2d53e36a3ae0f9bbd5f752acc00418d234ea08c5 100644
--- a/src/plugins/checkpoint/ompi/checkpoint_ompi.c
+++ b/src/plugins/checkpoint/ompi/checkpoint_ompi.c
@@ -243,35 +243,42 @@ extern int slurm_ckpt_free_job(check_jobinfo_t jobinfo)
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
struct check_job_info *check_ptr =
(struct check_job_info *)jobinfo;
- pack16(check_ptr->disabled, buffer);
- pack16(check_ptr->reply_cnt, buffer);
- pack16(check_ptr->wait_time, buffer);
+ if(protocol_version >= SLURM_2_1_PROTOCOL_VERSION) {
+ pack16(check_ptr->disabled, buffer);
+ pack16(check_ptr->reply_cnt, buffer);
+ pack16(check_ptr->wait_time, buffer);
- pack32(check_ptr->error_code, buffer);
- packstr(check_ptr->error_msg, buffer);
- pack_time(check_ptr->time_stamp, buffer);
+ pack32(check_ptr->error_code, buffer);
+ packstr(check_ptr->error_msg, buffer);
+ pack_time(check_ptr->time_stamp, buffer);
+ }
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
uint32_t uint32_tmp;
struct check_job_info *check_ptr =
(struct check_job_info *)jobinfo;
- safe_unpack16(&check_ptr->disabled, buffer);
- safe_unpack16(&check_ptr->reply_cnt, buffer);
- safe_unpack16(&check_ptr->wait_time, buffer);
+ if(protocol_version >= SLURM_2_1_PROTOCOL_VERSION) {
+ safe_unpack16(&check_ptr->disabled, buffer);
+ safe_unpack16(&check_ptr->reply_cnt, buffer);
+ safe_unpack16(&check_ptr->wait_time, buffer);
- safe_unpack32(&check_ptr->error_code, buffer);
- safe_unpackstr_xmalloc(&check_ptr->error_msg, &uint32_tmp, buffer);
- safe_unpack_time(&check_ptr->time_stamp, buffer);
+ safe_unpack32(&check_ptr->error_code, buffer);
+ safe_unpackstr_xmalloc(&check_ptr->error_msg,
+ &uint32_tmp, buffer);
+ safe_unpack_time(&check_ptr->time_stamp, buffer);
+ }
return SLURM_SUCCESS;
diff --git a/src/plugins/checkpoint/xlch/checkpoint_xlch.c b/src/plugins/checkpoint/xlch/checkpoint_xlch.c
index f0627a264a60c0c32a108e128a0ee2153cf6e27c..682723e6497b7d79016db5a6b3e00eb2fb3027ee 100644
--- a/src/plugins/checkpoint/xlch/checkpoint_xlch.c
+++ b/src/plugins/checkpoint/xlch/checkpoint_xlch.c
@@ -382,47 +382,54 @@ extern int slurm_ckpt_free_job(check_jobinfo_t jobinfo)
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_pack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
struct check_job_info *check_ptr =
(struct check_job_info *)jobinfo;
- pack16(check_ptr->disabled, buffer);
- pack16(check_ptr->task_cnt, buffer);
- pack16(check_ptr->reply_cnt, buffer);
- pack16(check_ptr->wait_time, buffer);
- pack_bit_fmt(check_ptr->replied, buffer);
+ if(protocol_version >= SLURM_2_1_PROTOCOL_VERSION) {
+ pack16(check_ptr->disabled, buffer);
+ pack16(check_ptr->task_cnt, buffer);
+ pack16(check_ptr->reply_cnt, buffer);
+ pack16(check_ptr->wait_time, buffer);
+ pack_bit_fmt(check_ptr->replied, buffer);
- pack32(check_ptr->error_code, buffer);
- packstr(check_ptr->error_msg, buffer);
- pack_time(check_ptr->time_stamp, buffer);
+ pack32(check_ptr->error_code, buffer);
+ packstr(check_ptr->error_msg, buffer);
+ pack_time(check_ptr->time_stamp, buffer);
+ }
return SLURM_SUCCESS;
}
-extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer)
+extern int slurm_ckpt_unpack_job(check_jobinfo_t jobinfo, Buf buffer,
+ uint16_t protocol_version)
{
uint32_t uint32_tmp;
char *task_inx_str;
struct check_job_info *check_ptr =
(struct check_job_info *)jobinfo;
- safe_unpack16(&check_ptr->disabled, buffer);
- safe_unpack16(&check_ptr->task_cnt, buffer);
- safe_unpack16(&check_ptr->reply_cnt, buffer);
- safe_unpack16(&check_ptr->wait_time, buffer);
- safe_unpackstr_xmalloc(&task_inx_str, &uint32_tmp, buffer);
- if (task_inx_str == NULL)
- check_ptr->replied = NULL;
- else {
- check_ptr->replied = bit_alloc(check_ptr->task_cnt);
- bit_unfmt(check_ptr->replied, task_inx_str);
- xfree(task_inx_str);
- }
+ if(protocol_version >= SLURM_2_1_PROTOCOL_VERSION) {
+ safe_unpack16(&check_ptr->disabled, buffer);
+ safe_unpack16(&check_ptr->task_cnt, buffer);
+ safe_unpack16(&check_ptr->reply_cnt, buffer);
+ safe_unpack16(&check_ptr->wait_time, buffer);
+ safe_unpackstr_xmalloc(&task_inx_str, &uint32_tmp, buffer);
+ if (task_inx_str == NULL)
+ check_ptr->replied = NULL;
+ else {
+ check_ptr->replied = bit_alloc(check_ptr->task_cnt);
+ bit_unfmt(check_ptr->replied, task_inx_str);
+ xfree(task_inx_str);
+ }
- safe_unpack32(&check_ptr->error_code, buffer);
- safe_unpackstr_xmalloc(&check_ptr->error_msg, &uint32_tmp, buffer);
- safe_unpack_time(&check_ptr->time_stamp, buffer);
+ safe_unpack32(&check_ptr->error_code, buffer);
+ safe_unpackstr_xmalloc(&check_ptr->error_msg,
+ &uint32_tmp, buffer);
+ safe_unpack_time(&check_ptr->time_stamp, buffer);
+ }
return SLURM_SUCCESS;
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 001ea978d12f5f8d98231630343af8b837ab8177..c3d796a4cd3693005035dde177ec583daa2547d4 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -782,7 +782,8 @@ static void _dump_job_state(struct job_record *dump_job_ptr, Buf buffer)
SLURM_PROTOCOL_VERSION);
pack16(dump_job_ptr->ckpt_interval, buffer);
- checkpoint_pack_jobinfo(dump_job_ptr->check_job, buffer);
+ checkpoint_pack_jobinfo(dump_job_ptr->check_job, buffer,
+ SLURM_PROTOCOL_VERSION);
packstr_array(dump_job_ptr->spank_job_env,
dump_job_ptr->spank_job_env_size, buffer);
@@ -946,7 +947,8 @@ static int _load_job_state(Buf buffer, uint16_t protocol_version)
safe_unpack16(&ckpt_interval, buffer);
if (checkpoint_alloc_jobinfo(&check_job) ||
- checkpoint_unpack_jobinfo(check_job, buffer))
+ checkpoint_unpack_jobinfo(check_job, buffer,
+ protocol_version))
goto unpack_error;
safe_unpackstr_array(&spank_job_env, &spank_job_env_size,
@@ -1091,7 +1093,8 @@ static int _load_job_state(Buf buffer, uint16_t protocol_version)
safe_unpack16(&ckpt_interval, buffer);
if (checkpoint_alloc_jobinfo(&check_job) ||
- checkpoint_unpack_jobinfo(check_job, buffer))
+ checkpoint_unpack_jobinfo(check_job, buffer,
+ protocol_version))
goto unpack_error;
safe_unpackstr_array(&spank_job_env, &spank_job_env_size,
diff --git a/src/slurmctld/step_mgr.c b/src/slurmctld/step_mgr.c
index 86069aa35c6c112ad79310de75e99a82b71247ef..3f0466a58a5e6a23e443c30b05389c137d7ef360 100644
--- a/src/slurmctld/step_mgr.c
+++ b/src/slurmctld/step_mgr.c
@@ -2483,7 +2483,8 @@ extern void dump_job_step_state(struct job_record *job_ptr,
SLURM_PROTOCOL_VERSION);
switch_pack_jobinfo(step_ptr->switch_job, buffer);
}
- checkpoint_pack_jobinfo(step_ptr->check_job, buffer);
+ checkpoint_pack_jobinfo(step_ptr->check_job, buffer,
+ SLURM_PROTOCOL_VERSION);
}
/*
@@ -2558,7 +2559,8 @@ extern int load_step_state(struct job_record *job_ptr, Buf buffer,
goto unpack_error;
}
checkpoint_alloc_jobinfo(&check_tmp);
- if (checkpoint_unpack_jobinfo(check_tmp, buffer))
+ if (checkpoint_unpack_jobinfo(check_tmp, buffer,
+ protocol_version))
goto unpack_error;
/* validity test as possible */
@@ -2614,7 +2616,8 @@ extern int load_step_state(struct job_record *job_ptr, Buf buffer,
goto unpack_error;
}
checkpoint_alloc_jobinfo(&check_tmp);
- if (checkpoint_unpack_jobinfo(check_tmp, buffer))
+ if (checkpoint_unpack_jobinfo(check_tmp, buffer,
+ protocol_version))
goto unpack_error;
/* validity test as possible */