int
svr_save_db(struct server *ps, int mode)
{
pbs_db_conn_t *conn = (pbs_db_conn_t *) svr_db_conn;
pbs_db_svr_info_t dbsvr;
pbs_db_obj_info_t obj;
int savetype = PBS_UPDATE_DB_FULL;
int rc;
ps->sv_qs.sv_savetm = time_now;
/* as part of the server save, update svrlive file now,
* used in failover
*/
if (update_svrlive() !=0)
return -1;
if (mode == SVR_SAVE_FULL)
savetype = PBS_UPDATE_DB_FULL;
else if (mode == SVR_SAVE_QUICK)
savetype = PBS_UPDATE_DB_QUICK;
else
savetype = PBS_INSERT_DB;
if (svr_to_db_svr(ps, &dbsvr, savetype) != 0)
goto db_err;
obj.pbs_db_obj_type = PBS_DB_SVR;
obj.pbs_db_un.pbs_db_svr = &dbsvr;
rc = pbs_db_save_obj(conn, &obj, savetype);
if (rc != 0) {
savetype = PBS_INSERT_DB;
rc = pbs_db_save_obj(conn, &obj, savetype);
}
pbs_db_reset_obj(&obj);
if (rc != 0)
goto db_err;
return (0);
db_err:
strcpy(log_buffer, msg_svdbnosv);
if (conn->conn_db_err != NULL)
strncat(log_buffer, conn->conn_db_err, LOG_BUF_SIZE - strlen(log_buffer) - 1);
log_err(-1, __func__, log_buffer);
panic_stop_db(log_buffer);
return (-1);
}
int
sched_save_db(pbs_sched *ps, int mode)
{
pbs_db_conn_t *conn = (pbs_db_conn_t *) svr_db_conn;
pbs_db_sched_info_t dbsched;
pbs_db_obj_info_t obj;
int savetype = PBS_UPDATE_DB_FULL;
int rc;
if (mode == SVR_SAVE_FULL)
savetype = PBS_UPDATE_DB_FULL;
else if (mode == SVR_SAVE_QUICK)
savetype = PBS_UPDATE_DB_QUICK;
else
savetype = PBS_INSERT_DB;
if (svr_to_db_sched(ps, &dbsched, savetype) != 0)
goto db_err;
obj.pbs_db_obj_type = PBS_DB_SCHED;
obj.pbs_db_un.pbs_db_sched = &dbsched;
rc = pbs_db_save_obj(conn, &obj, savetype);
if (rc != 0) {
savetype = PBS_INSERT_DB;
rc = pbs_db_save_obj(conn, &obj, savetype);
}
/* free the attribute list allocated by encode_attrs */
pbs_db_reset_obj(&obj);
if (rc != 0)
goto db_err;
return (0);
db_err:
strcpy(log_buffer, schedemsg);
if (conn->conn_db_err != NULL)
strncat(log_buffer, conn->conn_db_err, LOG_BUF_SIZE - strlen(log_buffer) - 1);
log_err(-1, __func__, log_buffer);
panic_stop_db(log_buffer);
return (-1);
}
/**
* @brief
* Save a queue to the database
*
* @param[in] pque - Pointer to the queue to save
* @param[in] mode:
* QUE_SAVE_FULL - Save full queue information (update)
* QUE_SAVE_NEW - Save new queue information (insert)
*
* @return Error code
* @retval 0 - Success
* @retval 1 - Failure
*
*/
int
que_save_db(pbs_queue *pque, int mode)
{
pbs_db_que_info_t dbque;
pbs_db_obj_info_t obj;
pbs_db_conn_t *conn = (pbs_db_conn_t *) svr_db_conn;
int savetype = PBS_UPDATE_DB_FULL;
if (svr_to_db_que(pque, &dbque, savetype) != 0)
goto db_err;
obj.pbs_db_obj_type = PBS_DB_QUEUE;
obj.pbs_db_un.pbs_db_que = &dbque;
if (mode == QUE_SAVE_NEW)
savetype = PBS_INSERT_DB;
if (pbs_db_save_obj(conn, &obj, savetype) != 0)
goto db_err;
pbs_db_reset_obj(&obj);
return (0);
db_err:
/* free the attribute list allocated by encode_attrs */
free(dbque.attr_list.attributes);
strcpy(log_buffer, "que_save failed ");
if (conn->conn_db_err != NULL)
strncat(log_buffer, conn->conn_db_err, LOG_BUF_SIZE - strlen(log_buffer) - 1);
log_err(-1, __func__, log_buffer);
panic_stop_db(log_buffer);
return (-1);
}
/**
* @brief
* Function to migrate filesystem data to database.
* Reads serverdb, scheddb, job files, node, nodestate, queue, resv information
* from the filesystem and save them into the database. All the information is
* recovered and saved into the database under a single database transaction,
* so any failure rolls back all the updates to the database. If all the updates
* to the database succeed, only then the respective files are deleted from the
* filesystem, else no deletion takes place.
*
* @return Error code
* @retval 0 : success
* @retval -1 : Failure
*
*/
int
svr_migrate_data_from_fs(void)
{
int baselen;
struct dirent *pdirent;
DIR *dir;
int had;
char *job_suffix = JOB_FILE_SUFFIX;
int job_suf_len = strlen(job_suffix);
job *pjob = NULL;
pbs_queue *pque;
resc_resv *presv;
char *psuffix;
int rc;
int recovered = 0;
char basen[MAXPATHLEN+1];
char scrfile[MAXPATHLEN+1];
char jobfile[MAXPATHLEN+1];
char origdir[MAXPATHLEN+1];
int fd;
struct stat stbuf;
char *scrbuf = NULL;
pbs_db_jobscr_info_t jobscr;
pbs_db_obj_info_t obj;
path_svrdb_new = build_path(path_priv, PBS_SERVERDB, new_tag);
path_scheddb = build_path(path_priv, PBS_SCHEDDB, NULL);
path_scheddb_new = build_path(path_priv, PBS_SCHEDDB, new_tag);
path_queues = build_path(path_priv, PBS_QUEDIR, suffix_slash);
path_resvs = build_path(path_priv, PBS_RESVDIR, suffix_slash);
path_nodes = build_path(path_priv, NODE_DESCRIP, NULL);
path_nodestate = build_path(path_priv, NODE_STATUS, NULL);
/* If not a "create" initialization, recover server db */
/* and sched db */
if (chk_save_file(path_svrdb) != 0) {
fprintf(stderr, "No serverdb found to update to datastore\n");
return (0);
}
if (setup_resc(1) == -1) {
fprintf(stderr, "%s\n", log_buffer);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
return (-1);
}
init_server_attrs();
/* start a database transation for the whole recovery */
if (pbs_db_begin_trx(svr_db_conn, 0, 0) != 0)
return (-1);
/* preprocess the nodes file to convert old properties to resources */
if (setup_nodes_fs(1) == -1) {
fprintf(stderr, "%s\n", log_buffer);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
return (-1);
}
/* Open the server database (save file) and read it in */
if (svr_recov_fs(path_svrdb) == -1) {
fprintf(stderr, "%s\n", msg_init_baddb);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
return (-1);
}
/* save server information to database now */
if (svr_save_db(&server, SVR_SAVE_NEW) != 0) {
fprintf(stderr, "Could not save server db\n");
if (svr_db_conn->conn_db_err)
fprintf(stderr, "[%s]\n", (char*)svr_db_conn->conn_db_err);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
return (-1);
}
/* now do sched db */
if (sched_recov_fs(path_scheddb) == -1) {
fprintf(stderr, "Unable to recover scheddb\n");
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
return (-1);
}
if (sched_save_db(dflt_scheduler, SVR_SAVE_NEW) != 0) {
fprintf(stderr, "Could not save scheduler db\n");
if (svr_db_conn->conn_db_err)
fprintf(stderr, "[%s]\n", (char*)svr_db_conn->conn_db_err);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
return (-1);
}
set_sched_default(dflt_scheduler, 0);
/* save current working dir before any chdirs */
if (getcwd(origdir, MAXPATHLEN) == NULL) {
fprintf(stderr, "getcwd failed\n");
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
return (-1);
}
if (chdir(path_queues) != 0) {
fprintf(stderr, msg_init_chdir, path_queues);
fprintf(stderr, "\n");
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
chdir(origdir);
return (-1);
}
had = server.sv_qs.sv_numque;
server.sv_qs.sv_numque = 0;
dir = opendir(".");
if (dir == NULL) {
fprintf(stderr, "%s\n", msg_init_noqueues);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
chdir(origdir);
return (-1);
}
while (errno = 0, (pdirent = readdir(dir)) != NULL) {
if (chk_save_file(pdirent->d_name) == 0) {
if ((pque = que_recov_fs(pdirent->d_name)) !=
NULL) {
/* que_recov increments sv_numque */
fprintf(stderr, msg_init_recovque,
pque->qu_qs.qu_name);
fprintf(stderr, "\n");
if (que_save_db(pque, QUE_SAVE_NEW) != 0) {
fprintf(stderr,
"Could not save queue info for queue %s\n",
pque->qu_qs.qu_name);
if (svr_db_conn->conn_db_err)
fprintf(stderr, "[%s]\n", (char*)svr_db_conn->conn_db_err);
(void) pbs_db_end_trx(svr_db_conn,
PBS_DB_ROLLBACK);
(void) closedir(dir);
chdir(origdir);
return (-1);
}
}
}
}
if (errno != 0 && errno != ENOENT) {
fprintf(stderr, "%s\n", msg_init_noqueues);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
(void) closedir(dir);
chdir(origdir);
return (-1);
}
(void) closedir(dir);
if (had != server.sv_qs.sv_numque) {
fprintf(stderr, msg_init_expctq, had, server.sv_qs.sv_numque);
fprintf(stderr, "\n");
}
/* Open and read in node list if one exists */
if (setup_nodes_fs(0) == -1) {
fprintf(stderr, "%s\n", log_buffer);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
chdir(origdir);
return (-1);
}
/*
* Recover reservations.
*/
if (chdir(path_resvs) != 0) {
fprintf(stderr, msg_init_chdir, path_resvs);
fprintf(stderr, "\n");
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
chdir(origdir);
return (-1);
}
dir = opendir(".");
if (dir == NULL) {
fprintf(stderr, "%s\n", msg_init_noresvs);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
chdir(origdir);
return (-1);
}
while (errno = 0, (pdirent = readdir(dir)) != NULL) {
if (chk_save_file(pdirent->d_name) == 0) {
presv = (resc_resv *)
job_or_resv_recov_fs(pdirent->d_name,
RESC_RESV_OBJECT);
if (presv != NULL) {
if (resv_save_db(presv, SAVERESV_NEW) != 0) {
fprintf(stderr,
"Could not save resv info for resv %s\n",
presv->ri_qs.ri_resvID);
if (svr_db_conn->conn_db_err)
fprintf(stderr, "[%s]\n", (char*)svr_db_conn->conn_db_err);
(void) pbs_db_end_trx(svr_db_conn,
PBS_DB_ROLLBACK);
(void) closedir(dir);
chdir(origdir);
return (-1);
}
}
}
}
if (errno != 0 && errno != ENOENT) {
fprintf(stderr, "%s\n", msg_init_noresvs);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
(void) closedir(dir);
chdir(origdir);
return (-1);
}
(void) closedir(dir);
/*
* Recover jobs
*/
if (chdir(path_jobs) != 0) {
fprintf(stderr, msg_init_chdir, path_jobs);
fprintf(stderr, "\n");
chdir(origdir);
return (-1);
}
server.sv_qs.sv_numjobs = 0;
recovered = 0;
dir = opendir(".");
if (dir == NULL) {
fprintf(stderr, "%s\n", msg_init_nojobs);
} else {
/* Now, for each job found ... */
while (errno = 0,
(pdirent = readdir(dir)) != NULL) {
if (chk_save_file(pdirent->d_name) != 0)
continue;
/* recover the job */
baselen = strlen(pdirent->d_name) - job_suf_len;
psuffix = pdirent->d_name + baselen;
if (strcmp(psuffix, job_suffix))
continue;
if ((pjob = job_recov_fs(pdirent->d_name)) == NULL) {
(void)strcpy(basen, pdirent->d_name);
psuffix = basen + baselen;
(void)strcpy(psuffix, JOB_BAD_SUFFIX);
(void)snprintf(log_buffer, sizeof(log_buffer), "moved bad file to %s",
basen);
log_event(PBSEVENT_SYSTEM,
PBS_EVENTCLASS_SERVER, LOG_NOTICE,
msg_daemonname, log_buffer);
continue;
}
if (pjob->ji_qs.ji_svrflags & JOB_SVFLG_SCRIPT) {
/* load the job script file */
strcpy(scrfile, path_jobs);
#ifndef WIN32
/* under WIN32, there's already a prefixed '/' */
(void) strcat(scrfile, "/");
#endif
strcat(scrfile, pdirent->d_name);
baselen = strlen(scrfile) - strlen(JOB_FILE_SUFFIX);
scrfile[baselen] = 0; /* put null char */
strcat(scrfile, JOB_SCRIPT_SUFFIX);
rc = 1;
#ifdef WIN32
if ((fd = open(scrfile, O_BINARY | O_RDONLY)) != -1)
#else
if ((fd = open(scrfile, O_RDONLY)) != -1)
#endif
{
/* load the script */
if (fstat(fd, &stbuf) == 0) {
if ((scrbuf = malloc(stbuf.st_size + 1))) {
if (read(fd, scrbuf, stbuf.st_size) == stbuf.st_size) {
scrbuf[stbuf.st_size] = '\0'; /* null character */
rc = 0; /* success loading */
}
}
}
close(fd);
}
if (rc != 0) {
fprintf(stderr, "Could not recover script file for job %s\n", pjob->ji_qs.ji_jobid);
(void) strcpy(basen, scrfile);
psuffix = basen + strlen(scrfile) - strlen(JOB_SCRIPT_SUFFIX);
(void) strcpy(psuffix, JOB_BAD_SUFFIX);
(void) strcpy(jobfile, scrfile);
psuffix = jobfile + strlen(jobfile) - strlen(JOB_SCRIPT_SUFFIX);
(void) strcpy(psuffix, JOB_FILE_SUFFIX);
#ifdef WIN32
if (MoveFileEx(jobfile, basen,
MOVEFILE_REPLACE_EXISTING | MOVEFILE_WRITE_THROUGH) == 0) {
errno = GetLastError();
snprintf(log_buffer, sizeof(log_buffer), "MoveFileEx(%s, %s) failed!",
jobfile, basen);
log_err(errno, "script", log_buffer);
}
secure_file(basen, "Administrators",
READS_MASK | WRITES_MASK | STANDARD_RIGHTS_REQUIRED);
#else
if (rename(jobfile, basen) == -1) {
snprintf(log_buffer, sizeof(log_buffer), "error renaming job file %s",
jobfile);
log_err(errno, "job_recov", log_buffer);
}
#endif
(void) snprintf(log_buffer, sizeof(log_buffer), "moved bad file to %s",
basen);
log_event(PBSEVENT_SYSTEM,
PBS_EVENTCLASS_SERVER, LOG_NOTICE,
msg_daemonname, log_buffer);
free(scrbuf);
scrbuf = NULL;
continue;
}
}
/* now save job first */
if (job_save_db(pjob, SAVEJOB_NEW) != 0) {
fprintf(stderr, "Could not save job info for jobid %s\n",
pjob->ji_qs.ji_jobid);
if (svr_db_conn->conn_db_err)
fprintf(stderr, "[%s]\n", (char*)svr_db_conn->conn_db_err);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
(void) closedir(dir);
chdir(origdir);
free(scrbuf);
return (-1);
}
if (pjob->ji_qs.ji_svrflags & JOB_SVFLG_SCRIPT) {
/* save job script */
strcpy(jobscr.ji_jobid, pjob->ji_qs.ji_jobid);
jobscr.script = scrbuf;
obj.pbs_db_obj_type = PBS_DB_JOBSCR;
obj.pbs_db_un.pbs_db_jobscr = &jobscr;
if (pbs_db_save_obj(svr_db_conn, &obj, PBS_INSERT_DB) != 0) {
fprintf(stderr, "Could not save job script for jobid %s\n",
pjob->ji_qs.ji_jobid);
if (svr_db_conn->conn_db_err)
fprintf(stderr, "[%s]\n", (char*)svr_db_conn->conn_db_err);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
free(scrbuf);
(void) closedir(dir);
chdir(origdir);
return (-1);
}
free(scrbuf);
scrbuf = NULL;
}
recovered++;
}
if (errno != 0 && errno != ENOENT) {
if (pjob)
fprintf(stderr, "readdir error for jobid %s\n", pjob->ji_qs.ji_jobid);
else
fprintf(stderr, "readdir error\n");
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
free(scrbuf);
(void) closedir(dir);
chdir(origdir);
return (-1);
}
(void) closedir(dir);
fprintf(stderr, msg_init_exptjobs, recovered);
fprintf(stderr, "\n");
}
if (save_nodes_db(0, NULL) != 0) {
fprintf(stderr, "Could not save nodes\n");
if (svr_db_conn->conn_db_err)
fprintf(stderr, "[%s]\n", (char*)svr_db_conn->conn_db_err);
(void) pbs_db_end_trx(svr_db_conn, PBS_DB_ROLLBACK);
chdir(origdir);
return (-1);
}
if (pbs_db_end_trx(svr_db_conn, PBS_DB_COMMIT) == 0) {
rm_migrated_files(path_priv);
chdir(origdir);
return (0);
}
chdir(origdir);
return -1;
}
