int makeflow_clean_file( struct dag *d, struct batch_queue *queue, struct dag_file *f, int silent)
{
if(!f)
return 1;
if(batch_fs_unlink(queue, f->filename) == 0) {
makeflow_log_file_state_change(d, f, DAG_FILE_STATE_DELETE);
debug(D_MAKEFLOW_RUN, "File deleted %s\n", f->filename);
} else if(errno != ENOENT) {
if(f->state == DAG_FILE_STATE_EXPECT || dag_file_should_exist(f))
makeflow_log_file_state_change(d, f, DAG_FILE_STATE_DELETE);
debug(D_MAKEFLOW_RUN, "Makeflow: Couldn't delete %s: %s\n", f->filename, strerror(errno));
return 1;
}
return 0;
}
static int node_submit( void * instance_struct, struct dag_node *n, struct batch_task *t){
struct vc3_definition *v = (struct vc3_definition*)instance_struct;
struct batch_wrapper *wrapper = batch_wrapper_create();
batch_wrapper_prefix(wrapper, "./vc3_builder_");
char * executable = NULL;
// If the queue supports remote_renaming add as remote rename.
if (batch_queue_supports_feature(makeflow_get_queue(n), "remote_rename")) {
executable = string_format("./%s", path_basename(v->exe));
} else {
// Else just use executable in path
executable = string_format("%s", v->exe);
}
/* Assumes a /disk dir in the image. */
char *log = string_format("%s_%d", v->log, t->taskid);
char *task_cmd = string_escape_shell(t->command);
char *cmd = string_format("%s --home $PWD %s -- %s > %s", executable, v->opt, task_cmd, log);
makeflow_hook_add_input_file(n->d, t, v->exe, executable, DAG_FILE_TYPE_GLOBAL);
makeflow_hook_add_output_file(n->d, t, log, log, DAG_FILE_TYPE_INTERMEDIATE);
free(log);
free(executable);
free(task_cmd);
batch_wrapper_cmd(wrapper, cmd);
free(cmd);
cmd = batch_wrapper_write(wrapper, t);
if(cmd){
batch_task_set_command(t, cmd);
struct dag_file *df = makeflow_hook_add_input_file(n->d, t, cmd, cmd, DAG_FILE_TYPE_TEMP);
debug(D_MAKEFLOW_HOOK, "Wrapper written to %s", df->filename);
makeflow_log_file_state_change(n->d, df, DAG_FILE_STATE_EXISTS);
} else {
debug(D_MAKEFLOW_HOOK, "Failed to create wrapper: errno %d, %s", errno, strerror(errno));
return MAKEFLOW_HOOK_FAILURE;
}
free(cmd);
return MAKEFLOW_HOOK_SUCCESS;
}
static int makeflow_module_sandbox_node_submit( void * instance_struct, struct dag_node *node, struct batch_task *task){
struct batch_wrapper *wrapper = batch_wrapper_create();
char *wrap_name = string_format("./task_%d_sandbox", task->taskid);
batch_wrapper_prefix(wrapper, wrap_name);
/* Save the directory we were originally working in. */
batch_wrapper_pre(wrapper, "export CUR_WORK_DIR=$(pwd)");
/* Create sandbox. This should probably have a hex or random tail to be unique. */
char *cmd = string_format("export SANDBOX=$(mktemp -d %s_XXXXXX)", wrap_name);
batch_wrapper_pre(wrapper, cmd);
free(cmd);
free(wrap_name);
struct batch_file *f;
list_first_item(task->input_files);
while((f = list_next_item(task->input_files))){
/* Skip if absolute path. */
if(f->inner_name[0] == '/') continue;
/* Add a cp for each file. Not linking as wq may already have done this. Not moving as it may be local. */
cmd = string_format("mkdir -p $(dirname $SANDBOX/%s) && cp -r %s $SANDBOX/%s", f->inner_name, f->inner_name, f->inner_name);
batch_wrapper_pre(wrapper, cmd);
free(cmd);
}
/* Enter into sandbox_dir. */
batch_wrapper_pre(wrapper, "cd $SANDBOX");
/* Execute the previous levels commmand. */
batch_wrapper_cmd(wrapper, task->command);
/* Once the command is finished go back to working dir. */
batch_wrapper_post(wrapper, "cd $CUR_WORK_DIR");
list_first_item(task->output_files);
while((f = list_next_item(task->output_files))){
/* Skip if absolute path. */
if(f->inner_name[0] == '/') continue;
/* Copy out results to expected location. OR TRUE so that lack of one file does not
prevent other files from being sent back.*/
cmd = string_format("mkdir -p $(dirname %s) && cp -r $SANDBOX/%s %s || true", f->inner_name, f->inner_name, f->inner_name);
batch_wrapper_post(wrapper, cmd);
free(cmd);
}
/* Remove and fully wipe out sandbox. */
batch_wrapper_post(wrapper, "rm -rf $SANDBOX");
cmd = batch_wrapper_write(wrapper, task);
if(cmd){
batch_task_set_command(task, cmd);
struct dag_file *df = makeflow_hook_add_input_file(node->d, task, cmd, cmd, DAG_FILE_TYPE_TEMP);
debug(D_MAKEFLOW_HOOK, "Wrapper written to %s", df->filename);
makeflow_log_file_state_change(node->d, df, DAG_FILE_STATE_EXISTS);
} else {
debug(D_MAKEFLOW_HOOK, "Failed to create wrapper: errno %d, %s", errno, strerror(errno));
return MAKEFLOW_HOOK_FAILURE;
}
free(cmd);
return MAKEFLOW_HOOK_SUCCESS;
}
/** The clean_mode variable was added so that we could better print out error messages
* apply in the situation. Currently only used to silence node rerun checking.
*/
void makeflow_log_recover(struct dag *d, const char *filename, int verbose_mode, struct batch_queue *queue, makeflow_clean_depth clean_mode)
{
char *line, *name, file[MAX_BUFFER_SIZE];
int nodeid, state, jobid, file_state;
int first_run = 1;
struct dag_node *n;
struct dag_file *f;
struct stat buf;
timestamp_t previous_completion_time;
d->logfile = fopen(filename, "r");
if(d->logfile) {
int linenum = 0;
first_run = 0;
printf("recovering from log file %s...\n",filename);
while((line = get_line(d->logfile))) {
linenum++;
if(sscanf(line, "# %d %s %" SCNu64 "", &file_state, file, &previous_completion_time) == 3) {
f = dag_file_lookup_or_create(d, file);
f->state = file_state;
if(file_state == DAG_FILE_STATE_EXISTS){
d->completed_files += 1;
f->creation_logged = (time_t) (previous_completion_time / 1000000);
} else if(file_state == DAG_FILE_STATE_DELETE){
d->deleted_files += 1;
}
continue;
}
if(line[0] == '#')
continue;
if(sscanf(line, "%" SCNu64 " %d %d %d", &previous_completion_time, &nodeid, &state, &jobid) == 4) {
n = itable_lookup(d->node_table, nodeid);
if(n) {
n->state = state;
n->jobid = jobid;
/* Log timestamp is in microseconds, we need seconds for diff. */
n->previous_completion = (time_t) (previous_completion_time / 1000000);
continue;
}
}
fprintf(stderr, "makeflow: %s appears to be corrupted on line %d\n", filename, linenum);
exit(1);
}
fclose(d->logfile);
}
d->logfile = fopen(filename, "a");
if(!d->logfile) {
fprintf(stderr, "makeflow: couldn't open logfile %s: %s\n", filename, strerror(errno));
exit(1);
}
if(setvbuf(d->logfile, NULL, _IOLBF, BUFSIZ) != 0) {
fprintf(stderr, "makeflow: couldn't set line buffer on logfile %s: %s\n", filename, strerror(errno));
exit(1);
}
if(first_run && verbose_mode) {
struct dag_file *f;
struct dag_node *p;
for(n = d->nodes; n; n = n->next) {
/* Record node information to log */
fprintf(d->logfile, "# NODE\t%d\t%s\n", n->nodeid, n->command);
/* Record the node category to the log */
fprintf(d->logfile, "# CATEGORY\t%d\t%s\n", n->nodeid, n->category->label);
fprintf(d->logfile, "# SYMBOL\t%d\t%s\n", n->nodeid, n->category->label); /* also write the SYMBOL as alias of CATEGORY, deprecated. */
/* Record node parents to log */
fprintf(d->logfile, "# PARENTS\t%d", n->nodeid);
list_first_item(n->source_files);
while( (f = list_next_item(n->source_files)) ) {
p = f->created_by;
if(p)
fprintf(d->logfile, "\t%d", p->nodeid);
}
fputc('\n', d->logfile);
/* Record node inputs to log */
fprintf(d->logfile, "# SOURCES\t%d", n->nodeid);
list_first_item(n->source_files);
while( (f = list_next_item(n->source_files)) ) {
fprintf(d->logfile, "\t%s", f->filename);
}
fputc('\n', d->logfile);
/* Record node outputs to log */
fprintf(d->logfile, "# TARGETS\t%d", n->nodeid);
list_first_item(n->target_files);
while( (f = list_next_item(n->target_files)) ) {
fprintf(d->logfile, "\t%s", f->filename);
}
fputc('\n', d->logfile);
/* Record translated command to log */
fprintf(d->logfile, "# COMMAND\t%d\t%s\n", n->nodeid, n->command);
}
}
dag_count_states(d);
// Check for log consistency
if(!first_run) {
hash_table_firstkey(d->files);
while(hash_table_nextkey(d->files, &name, (void **) &f)) {
if(dag_file_should_exist(f) && !dag_file_is_source(f) && !(batch_fs_stat(queue, f->filename, &buf) >= 0)){
fprintf(stderr, "makeflow: %s is reported as existing, but does not exist.\n", f->filename);
makeflow_log_file_state_change(d, f, DAG_FILE_STATE_UNKNOWN);
continue;
}
if(S_ISDIR(buf.st_mode))
continue;
if(dag_file_should_exist(f) && !dag_file_is_source(f) && difftime(buf.st_mtime, f->creation_logged) > 0) {
fprintf(stderr, "makeflow: %s is reported as existing, but has been modified (%" SCNu64 " ,%" SCNu64 ").\n", f->filename, (uint64_t)buf.st_mtime, (uint64_t)f->creation_logged);
makeflow_clean_file(d, queue, f, 0);
makeflow_log_file_state_change(d, f, DAG_FILE_STATE_UNKNOWN);
}
}
}
int silent = 0;
if(clean_mode != MAKEFLOW_CLEAN_NONE)
silent = 1;
// Decide rerun tasks
if(!first_run) {
struct itable *rerun_table = itable_create(0);
for(n = d->nodes; n; n = n->next) {
makeflow_node_decide_rerun(rerun_table, d, n, silent);
}
itable_delete(rerun_table);
}
//Update file reference counts from nodes in log
for(n = d->nodes; n; n = n->next) {
if(n->state == DAG_NODE_STATE_COMPLETE)
{
struct dag_file *f;
list_first_item(n->source_files);
while((f = list_next_item(n->source_files)))
f->ref_count += -1;
}
}
}
