/**
* @brief
* issue a batch request to another server or to a MOM
* or even to ourself!
*
* If the request is meant for this every server, then
* Set up work-task of type WORK_Deferred_Local with a dummy
* connection handle (PBS_LOCAL_CONNECTION).
*
* Dispatch the request to be processed. [reply_send() will
* dispatch the reply via the work task entry.]
*
* If the request is to another server/MOM, then
* Set up work-task of type WORK_Deferred_Reply with the
* connection handle as the event.
*
* Encode and send the request.
*
* When the reply is ready, process_reply() will decode it and
* dispatch the work task.
*
* @note
* IT IS UP TO THE FUNCTION DISPATCHED BY THE WORK TASK TO CLOSE THE
* CONNECTION (connection handle not socket) and FREE THE REQUEST
* STRUCTURE. The connection (non-negative if open) is in wt_event
* and the pointer to the request structure is in wt_parm1.
*
* @param[in] conn - connection index
* @param[in] request - batch request to send
* @param[in] func - The callback function to invoke to handle the batch reply
* @param[out] ppwt - Return work task to be maintained by server to handle
* deferred replies
* @param[in] rpp - conn is over tcp or rpp?
*
* @return Error code
* @retval 0 - Success
* @retval -1 - Failure
*
*/
int
issue_Drequest(int conn,
struct batch_request *request,
void (*func)(),
struct work_task **ppwt,
int rpp)
{
struct attropl *patrl;
struct work_task *ptask;
struct svrattrl *psvratl;
int rc;
int sock = -1;
enum work_type wt;
char *msgid = NULL;
request->rppcmd_msgid = NULL;
if (conn == PBS_LOCAL_CONNECTION) {
wt = WORK_Deferred_Local;
request->rq_conn = PBS_LOCAL_CONNECTION;
} else if (rpp) {
sock = conn;
request->rq_conn = conn;
wt = WORK_Deferred_Reply;
} else {
sock = connection[conn].ch_socket;
request->rq_conn = sock;
wt = WORK_Deferred_Reply;
DIS_tcp_setup(sock);
}
ptask = set_task(wt, (long) conn, func, (void *) request);
if (ptask == NULL) {
log_err(errno, __func__, "could not set_task");
if (ppwt != 0)
*ppwt = 0;
return (-1);
}
if (conn == PBS_LOCAL_CONNECTION) {
/* the request should be issued to ourself */
dispatch_request(PBS_LOCAL_CONNECTION, request);
if (ppwt != 0)
*ppwt = ptask;
return (0);
}
/* the request is bound to another server, encode/send the request */
switch (request->rq_type) {
#ifndef PBS_MOM
case PBS_BATCH_DeleteJob:
rc = PBSD_mgr_put(conn,
PBS_BATCH_DeleteJob,
MGR_CMD_DELETE,
MGR_OBJ_JOB,
request->rq_ind.rq_delete.rq_objname,
NULL,
request->rq_extend,
rpp,
&msgid);
break;
case PBS_BATCH_HoldJob:
attrl_fixlink(&request->rq_ind.rq_hold.rq_orig.rq_attr);
psvratl = (struct svrattrl *)GET_NEXT(
request->rq_ind.rq_hold.rq_orig.rq_attr);
patrl = &psvratl->al_atopl;
rc = PBSD_mgr_put(conn,
PBS_BATCH_HoldJob,
MGR_CMD_SET,
MGR_OBJ_JOB,
request->rq_ind.rq_hold.rq_orig.rq_objname,
patrl,
NULL,
rpp,
&msgid);
break;
case PBS_BATCH_MessJob:
rc = PBSD_msg_put(conn,
request->rq_ind.rq_message.rq_jid,
request->rq_ind.rq_message.rq_file,
request->rq_ind.rq_message.rq_text,
NULL,
rpp,
&msgid);
break;
case PBS_BATCH_RelnodesJob:
rc = PBSD_relnodes_put(conn,
request->rq_ind.rq_relnodes.rq_jid,
request->rq_ind.rq_relnodes.rq_node_list,
NULL,
rpp,
&msgid);
break;
case PBS_BATCH_PySpawn:
rc = PBSD_py_spawn_put(conn,
request->rq_ind.rq_py_spawn.rq_jid,
request->rq_ind.rq_py_spawn.rq_argv,
request->rq_ind.rq_py_spawn.rq_envp,
rpp,
&msgid);
break;
case PBS_BATCH_ModifyJob:
attrl_fixlink(&request->rq_ind.rq_modify.rq_attr);
patrl = (struct attropl *)&((struct svrattrl *)GET_NEXT(
request->rq_ind.rq_modify.rq_attr))->al_atopl;
rc = PBSD_mgr_put(conn,
PBS_BATCH_ModifyJob,
MGR_CMD_SET,
MGR_OBJ_JOB,
request->rq_ind.rq_modify.rq_objname,
patrl,
NULL,
rpp,
&msgid);
break;
case PBS_BATCH_Rerun:
if (rpp) {
rc = is_compose_cmd(sock, IS_CMD, &msgid);
if (rc != 0)
break;
}
rc=encode_DIS_ReqHdr(sock, PBS_BATCH_Rerun, pbs_current_user);
if (rc != 0)
break;
rc=encode_DIS_JobId(sock, request->rq_ind.rq_rerun);
if (rc != 0)
break;
rc=encode_DIS_ReqExtend(sock, 0);
if (rc != 0)
break;
rc = DIS_wflush(sock, rpp);
break;
case PBS_BATCH_RegistDep:
if (rpp) {
rc = is_compose_cmd(sock, IS_CMD, &msgid);
if (rc != 0)
break;
}
rc=encode_DIS_ReqHdr(sock,
PBS_BATCH_RegistDep, pbs_current_user);
if (rc != 0)
break;
rc=encode_DIS_Register(sock, request);
if (rc != 0)
break;
rc=encode_DIS_ReqExtend(sock, 0);
if (rc != 0)
break;
rc = DIS_wflush(sock, rpp);
break;
case PBS_BATCH_SignalJob:
rc = PBSD_sig_put(conn,
request->rq_ind.rq_signal.rq_jid,
request->rq_ind.rq_signal.rq_signame,
NULL,
rpp,
&msgid);
break;
case PBS_BATCH_StatusJob:
rc = PBSD_status_put(conn,
PBS_BATCH_StatusJob,
request->rq_ind.rq_status.rq_id,
NULL, NULL,
rpp,
&msgid);
break;
case PBS_BATCH_TrackJob:
if (rpp) {
rc = is_compose_cmd(sock, IS_CMD, &msgid);
if (rc != 0)
break;
}
rc=encode_DIS_ReqHdr(sock, PBS_BATCH_TrackJob, pbs_current_user);
if (rc != 0)
break;
rc=encode_DIS_TrackJob(sock, request);
if (rc != 0)
break;
rc=encode_DIS_ReqExtend(sock, request->rq_extend);
if (rc != 0)
break;
rc = DIS_wflush(sock, rpp);
break;
case PBS_BATCH_CopyFiles:
if (rpp) {
rc = is_compose_cmd(sock, IS_CMD, &msgid);
if (rc != 0)
break;
}
rc=encode_DIS_ReqHdr(sock,
PBS_BATCH_CopyFiles, pbs_current_user);
if (rc != 0)
break;
rc=encode_DIS_CopyFiles(sock, request);
if (rc != 0)
break;
rc=encode_DIS_ReqExtend(sock, 0);
if (rc != 0)
break;
rc = DIS_wflush(sock, rpp);
break;
case PBS_BATCH_CopyFiles_Cred:
if (rpp) {
rc = is_compose_cmd(sock, IS_CMD, &msgid);
if (rc != 0)
break;
}
rc=encode_DIS_ReqHdr(sock,
PBS_BATCH_CopyFiles_Cred, pbs_current_user);
if (rc != 0)
break;
rc=encode_DIS_CopyFiles_Cred(sock, request);
if (rc != 0)
break;
rc=encode_DIS_ReqExtend(sock, 0);
if (rc != 0)
break;
rc = DIS_wflush(sock, rpp);
break;
case PBS_BATCH_DelFiles:
if (rpp) {
rc = is_compose_cmd(sock, IS_CMD, &msgid);
if (rc != 0)
break;
}
rc=encode_DIS_ReqHdr(sock,
PBS_BATCH_DelFiles, pbs_current_user);
if (rc != 0)
break;
rc=encode_DIS_CopyFiles(sock, request);
if (rc != 0)
break;
rc=encode_DIS_ReqExtend(sock, 0);
if (rc != 0)
break;
rc = DIS_wflush(sock, rpp);
break;
case PBS_BATCH_DelFiles_Cred:
if (rpp) {
rc = is_compose_cmd(sock, IS_CMD, &msgid);
if (rc != 0)
break;
}
rc=encode_DIS_ReqHdr(sock,
PBS_BATCH_DelFiles_Cred, pbs_current_user);
if (rc != 0)
break;
rc=encode_DIS_CopyFiles_Cred(sock, request);
if (rc != 0)
break;
rc=encode_DIS_ReqExtend(sock, 0);
if (rc != 0)
break;
rc = DIS_wflush(sock, rpp);
break;
case PBS_BATCH_FailOver:
rc = put_failover(sock, request); /* we should never do rpp for this one */
break;
#else /* PBS_MOM */
case PBS_BATCH_JobObit:
rc=encode_DIS_ReqHdr(sock, PBS_BATCH_JobObit, pbs_current_user);
if (rc != 0)
break;
rc=encode_DIS_JobObit(sock, request);
if (rc != 0)
break;
rc=encode_DIS_ReqExtend(sock, 0);
if (rc != 0)
break;
rc = DIS_wflush(sock, rpp);
break;
#endif /* PBS_MOM */
default:
(void)sprintf(log_buffer, msg_issuebad, request->rq_type);
log_err(-1, __func__, log_buffer);
delete_task(ptask);
rc = -1;
break;
}
if (rc) {
sprintf(log_buffer,
"issue_Drequest failed, error=%d on request %d",
rc, request->rq_type);
log_err(-1, __func__, log_buffer);
if (msgid)
free(msgid);
delete_task(ptask);
} else if (ppwt != 0) {
if (rpp) {
rpp_add_close_func(sock, process_DreplyRPP); /* register a close handler */
ptask->wt_event2 = msgid;
/*
* since its an rpp delayed task, remove it from the task_event list
* caller will add to moms deferred cmd list
*/
delete_link(&ptask->wt_linkall);
}
ptask->wt_aux2 = rpp; /* 0 in case of non-TPP */
*ppwt = ptask;
}
return (rc);
}
int send_request_to_remote_server(
int conn,
batch_request *request)
{
struct attropl *patrl;
struct svrattrl *psvratl;
int rc = PBSE_NONE;
int tmp_rc = PBSE_NONE;
int sock = 0;
char log_buf[LOCAL_LOG_BUF_SIZE];
struct tcp_chan *chan = NULL;
pthread_mutex_lock(connection[conn].ch_mutex);
sock = connection[conn].ch_socket;
pthread_mutex_unlock(connection[conn].ch_mutex);
request->rq_conn = sock;
if ((chan = DIS_tcp_setup(sock)) == NULL)
{
log_err(PBSE_MEM_MALLOC, __func__,
"Could not allocate memory for socket buffer");
close_conn(sock, FALSE);
return(PBSE_MEM_MALLOC);
}
/* the request is bound to another server, encode/send the request */
switch (request->rq_type)
{
case PBS_BATCH_DeleteJob:
rc = PBSD_mgr_put(
conn,
PBS_BATCH_DeleteJob,
MGR_CMD_DELETE,
MGR_OBJ_JOB,
request->rq_ind.rq_delete.rq_objname,
NULL,
NULL);
break;
case PBS_BATCH_HoldJob:
attrl_fixlink(&request->rq_ind.rq_hold.rq_orig.rq_attr);
psvratl = (struct svrattrl *)GET_NEXT(request->rq_ind.rq_hold.rq_orig.rq_attr);
patrl = &psvratl->al_atopl;
rc = PBSD_mgr_put(
conn,
PBS_BATCH_HoldJob,
MGR_CMD_SET,
MGR_OBJ_JOB,
request->rq_ind.rq_hold.rq_orig.rq_objname,
patrl,
NULL);
break;
case PBS_BATCH_CheckpointJob:
rc = PBSD_mgr_put(
conn,
PBS_BATCH_CheckpointJob,
MGR_CMD_SET,
MGR_OBJ_JOB,
request->rq_ind.rq_hold.rq_orig.rq_objname,
NULL,
NULL);
break;
case PBS_BATCH_GpuCtrl:
rc = PBSD_gpu_put(
conn,
request->rq_ind.rq_gpuctrl.rq_momnode,
request->rq_ind.rq_gpuctrl.rq_gpuid,
request->rq_ind.rq_gpuctrl.rq_gpumode,
request->rq_ind.rq_gpuctrl.rq_reset_perm,
request->rq_ind.rq_gpuctrl.rq_reset_vol,
NULL);
break;
case PBS_BATCH_MessJob:
rc = PBSD_msg_put(
conn,
request->rq_ind.rq_message.rq_jid,
request->rq_ind.rq_message.rq_file,
request->rq_ind.rq_message.rq_text,
NULL);
break;
case PBS_BATCH_ModifyJob:
case PBS_BATCH_AsyModifyJob:
attrl_fixlink(&request->rq_ind.rq_modify.rq_attr);
patrl = (struct attropl *) & ((struct svrattrl *)GET_NEXT(
request->rq_ind.rq_modify.rq_attr))->al_atopl;
rc = PBSD_mgr_put(
conn,
request->rq_type,
MGR_CMD_SET,
MGR_OBJ_JOB,
request->rq_ind.rq_modify.rq_objname,
patrl,
NULL);
break;
case PBS_BATCH_Rerun:
if ((rc = encode_DIS_ReqHdr(chan, PBS_BATCH_Rerun, msg_daemonname)))
break;
if ((rc = encode_DIS_JobId(chan, request->rq_ind.rq_rerun)))
break;
if ((rc = encode_DIS_ReqExtend(chan, 0)))
break;
rc = DIS_tcp_wflush(chan);
break;
case PBS_BATCH_RegistDep:
if ((rc = encode_DIS_ReqHdr(chan, PBS_BATCH_RegistDep, msg_daemonname)))
break;
if ((rc = encode_DIS_Register(chan, request)))
break;
if ((rc = encode_DIS_ReqExtend(chan, 0)))
break;
rc = DIS_tcp_wflush(chan);
break;
case PBS_BATCH_AsySignalJob:
case PBS_BATCH_SignalJob:
rc = PBSD_sig_put(
conn,
request->rq_ind.rq_signal.rq_jid,
request->rq_ind.rq_signal.rq_signame,
request->rq_extra);
break;
case PBS_BATCH_StatusJob:
rc = PBSD_status_put(
conn,
PBS_BATCH_StatusJob,
request->rq_ind.rq_status.rq_id,
NULL,
NULL);
break;
case PBS_BATCH_TrackJob:
if ((rc = encode_DIS_ReqHdr(chan, PBS_BATCH_TrackJob, msg_daemonname)))
break;
if ((rc = encode_DIS_TrackJob(chan, request)))
break;
if ((rc = encode_DIS_ReqExtend(chan, 0)))
break;
rc = DIS_tcp_wflush(chan);
break;
case PBS_BATCH_ReturnFiles:
if ((rc = encode_DIS_ReqHdr(chan, PBS_BATCH_ReturnFiles, msg_daemonname)))
break;
if ((rc = encode_DIS_ReturnFiles(chan, request)))
break;
if ((rc = encode_DIS_ReqExtend(chan, 0)))
break;
rc = DIS_tcp_wflush(chan);
break;
case PBS_BATCH_CopyFiles:
if ((rc = encode_DIS_ReqHdr(chan, PBS_BATCH_CopyFiles, msg_daemonname)))
break;
if ((rc = encode_DIS_CopyFiles(chan, request)))
break;
if ((rc = encode_DIS_ReqExtend(chan, 0)))
break;
rc = DIS_tcp_wflush(chan);
break;
case PBS_BATCH_DelFiles:
if ((rc = encode_DIS_ReqHdr(chan, PBS_BATCH_DelFiles, msg_daemonname)))
break;
if ((rc = encode_DIS_CopyFiles(chan, request)))
break;
if ((rc = encode_DIS_ReqExtend(chan, 0)))
break;
rc = DIS_tcp_wflush(chan);
break;
case PBS_BATCH_DeleteReservation:
if ((rc = encode_DIS_ReqHdr(chan, PBS_BATCH_DeleteReservation, msg_daemonname)))
break;
if ((rc = encode_DIS_ReqExtend(chan, request->rq_extend)))
break;
rc = DIS_tcp_wflush(chan);
break;
default:
sprintf(log_buf, msg_issuebad, request->rq_type);
log_err(-1, __func__, log_buf);
rc = -1;
break;
} /* END switch (request->rq_type) */
if ((tmp_rc = DIS_reply_read(chan, &request->rq_reply)) != 0)
{
sprintf(log_buf, "DIS_reply_read failed: %d", tmp_rc);
log_record(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
request->rq_reply.brp_code = tmp_rc;
request->rq_reply.brp_choice = BATCH_REPLY_CHOICE_NULL;
}
DIS_tcp_cleanup(chan);
svr_disconnect(conn);
return(rc);
} /* END send_request_to_remote_server() */
int issue_Drequest(
int conn,
struct batch_request *request,
void (*func) (struct work_task *),
struct work_task **ppwt)
{
struct attropl *patrl;
struct work_task *ptask;
struct svrattrl *psvratl;
int rc;
int sock = 0;
enum work_type wt;
char *id = "issue_Drequest";
if (conn == PBS_LOCAL_CONNECTION)
{
wt = WORK_Deferred_Local;
request->rq_conn = PBS_LOCAL_CONNECTION;
}
else
{
sock = connection[conn].ch_socket;
request->rq_conn = sock;
wt = WORK_Deferred_Reply;
DIS_tcp_setup(sock);
}
ptask = set_task(wt, (long)conn, func, (void *)request);
if (ptask == NULL)
{
log_err(errno, id, "could not set_task");
if (ppwt != NULL)
*ppwt = 0;
return(-1);
}
if (conn == PBS_LOCAL_CONNECTION)
{
/* the request should be issued to ourself */
dispatch_request(PBS_LOCAL_CONNECTION, request);
if (ppwt != NULL)
*ppwt = ptask;
return(0);
}
/* the request is bound to another server, encode/send the request */
switch (request->rq_type)
{
#ifndef PBS_MOM
case PBS_BATCH_DeleteJob:
rc = PBSD_mgr_put(
conn,
PBS_BATCH_DeleteJob,
MGR_CMD_DELETE,
MGR_OBJ_JOB,
request->rq_ind.rq_delete.rq_objname,
NULL,
NULL);
break;
case PBS_BATCH_HoldJob:
attrl_fixlink(&request->rq_ind.rq_hold.rq_orig.rq_attr);
psvratl = (struct svrattrl *)GET_NEXT(request->rq_ind.rq_hold.rq_orig.rq_attr);
patrl = &psvratl->al_atopl;
rc = PBSD_mgr_put(
conn,
PBS_BATCH_HoldJob,
MGR_CMD_SET,
MGR_OBJ_JOB,
request->rq_ind.rq_hold.rq_orig.rq_objname,
patrl,
NULL);
break;
case PBS_BATCH_CheckpointJob:
rc = PBSD_mgr_put(
conn,
PBS_BATCH_CheckpointJob,
MGR_CMD_SET,
MGR_OBJ_JOB,
request->rq_ind.rq_hold.rq_orig.rq_objname,
NULL,
NULL);
break;
case PBS_BATCH_GpuCtrl:
rc = PBSD_gpu_put(
conn,
request->rq_ind.rq_gpuctrl.rq_momnode,
request->rq_ind.rq_gpuctrl.rq_gpuid,
request->rq_ind.rq_gpuctrl.rq_gpumode,
request->rq_ind.rq_gpuctrl.rq_reset_perm,
request->rq_ind.rq_gpuctrl.rq_reset_vol,
NULL);
break;
case PBS_BATCH_MessJob:
rc = PBSD_msg_put(
conn,
request->rq_ind.rq_message.rq_jid,
request->rq_ind.rq_message.rq_file,
request->rq_ind.rq_message.rq_text,
NULL);
break;
case PBS_BATCH_ModifyJob:
case PBS_BATCH_AsyModifyJob:
attrl_fixlink(&request->rq_ind.rq_modify.rq_attr);
patrl = (struct attropl *) & ((struct svrattrl *)GET_NEXT(
request->rq_ind.rq_modify.rq_attr))->al_atopl;
rc = PBSD_mgr_put(
conn,
request->rq_type,
MGR_CMD_SET,
MGR_OBJ_JOB,
request->rq_ind.rq_modify.rq_objname,
patrl,
NULL);
break;
case PBS_BATCH_Rerun:
if ((rc = encode_DIS_ReqHdr(sock, PBS_BATCH_Rerun, msg_daemonname)))
break;
if ((rc = encode_DIS_JobId(sock, request->rq_ind.rq_rerun)))
break;
if ((rc = encode_DIS_ReqExtend(sock, 0)))
break;
rc = DIS_tcp_wflush(sock);
break;
case PBS_BATCH_RegistDep:
if ((rc = encode_DIS_ReqHdr(sock, PBS_BATCH_RegistDep, msg_daemonname)))
break;
if ((rc = encode_DIS_Register(sock, request)))
break;
if ((rc = encode_DIS_ReqExtend(sock, 0)))
break;
rc = DIS_tcp_wflush(sock);
break;
case PBS_BATCH_AsySignalJob:
case PBS_BATCH_SignalJob:
rc = PBSD_sig_put(
conn,
request->rq_ind.rq_signal.rq_jid,
request->rq_ind.rq_signal.rq_signame,
request->rq_extra);
break;
case PBS_BATCH_StatusJob:
rc = PBSD_status_put(
conn,
PBS_BATCH_StatusJob,
request->rq_ind.rq_status.rq_id,
NULL,
NULL);
break;
case PBS_BATCH_TrackJob:
if ((rc = encode_DIS_ReqHdr(sock, PBS_BATCH_TrackJob, msg_daemonname)))
break;
if ((rc = encode_DIS_TrackJob(sock, request)))
break;
if ((rc = encode_DIS_ReqExtend(sock, 0)))
break;
rc = DIS_tcp_wflush(sock);
break;
case PBS_BATCH_ReturnFiles:
if ((rc = encode_DIS_ReqHdr(sock, PBS_BATCH_ReturnFiles, msg_daemonname)))
break;
if ((rc = encode_DIS_ReturnFiles(sock, request)))
break;
if ((rc = encode_DIS_ReqExtend(sock, 0)))
break;
rc = DIS_tcp_wflush(sock);
break;
case PBS_BATCH_CopyFiles:
if ((rc = encode_DIS_ReqHdr(sock, PBS_BATCH_CopyFiles, msg_daemonname)))
break;
if ((rc = encode_DIS_CopyFiles(sock, request)))
break;
if ((rc = encode_DIS_ReqExtend(sock, 0)))
break;
rc = DIS_tcp_wflush(sock);
break;
case PBS_BATCH_DelFiles:
if ((rc = encode_DIS_ReqHdr(sock, PBS_BATCH_DelFiles, msg_daemonname)))
break;
if ((rc = encode_DIS_CopyFiles(sock, request)))
break;
if ((rc = encode_DIS_ReqExtend(sock, 0)))
break;
rc = DIS_tcp_wflush(sock);
break;
#else /* PBS_MOM */
case PBS_BATCH_JobObit:
/* who is sending obit request? */
if ((rc = encode_DIS_ReqHdr(sock, PBS_BATCH_JobObit, msg_daemonname)))
break;
if ((rc = encode_DIS_JobObit(sock, request)))
break;
if ((rc = encode_DIS_ReqExtend(sock, 0)))
break;
rc = DIS_tcp_wflush(sock);
break;
#endif /* PBS_MOM */
default:
sprintf(log_buffer, msg_issuebad,
request->rq_type);
log_err(-1, id, log_buffer);
delete_task(ptask);
rc = -1;
break;
} /* END switch (request->rq_type) */
if (ppwt != NULL)
*ppwt = ptask;
return(rc);
} /* END issue_Drequest() */