Skip to content

Commit

Permalink
Merge pull request #5938 from BOINC/dpa_start
Browse files Browse the repository at this point in the history 
fix problem where wrapper unzips input files each time it restarts
  • Loading branch information
@AenBleidd
AenBleidd authored Dec 7, 2024
2 parents 51803c6 + 1ac38c9 commit 54266b2
Show file tree
Hide file tree
Showing 7 changed files with 130 additions and 183 deletions.
3 changes: 2 additions & 1 deletion client/app.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -233,7 +233,8 @@ struct ACTIVE_TASK {
// This is compared with project-specified limits
// to decide whether to abort job; no other use.
int get_free_slot(RESULT*);
int start(bool test=false); // start a process
int setup_slot_dir(char *buf, unsigned int size);
int start();

// Termination stuff.
// Terminology:
Expand Down
5 changes: 0 additions & 5 deletions client/app_control.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -596,11 +596,6 @@ void ACTIVE_TASK::handle_exited_app(int stat) {
//
cleanup_task();

if (gstate.run_test_app) {
msg_printf(0, MSG_INFO, "test app finished - exiting");
exit(0);
}

if (!will_restart) {
copy_output_files();
int retval = read_stderr_file();
Expand Down
283 changes: 120 additions & 163 deletions client/app_start.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -521,9 +521,100 @@ static int get_priority(bool is_high_priority) {
#endif
}

// set up slot dir; copy or link app version and input files.
// If copy a big file, setup_file() will start an async copy
// and return ERR_IN_PROGRESS.
//
int ACTIVE_TASK::setup_slot_dir(char *buf, unsigned int buf_len) {
unsigned int i;
int retval;
FILE_REF fref;
FILE_INFO *fip;
char file_path[MAXPATHLEN];
char path[1024];

*buf = 0;

sprintf(path, "%s/boinc_setup_complete", slot_dir);
if (boinc_file_exists(path)) return 0;

// make sure the needed files exist
//
retval = gstate.task_files_present(result, true, &fip);
if (retval) {
snprintf(
buf, buf_len,
"Task file %s: %s",
fip->name, boincerror(retval)
);
return retval;
}

// set up app version files
//
for (i=0; i<app_version->app_files.size(); i++) {
fref = app_version->app_files[i];
fip = fref.file_info;
get_pathname(fip, file_path, sizeof(file_path));
if (fref.main_program) {
if (is_image_file(fip->name)) {
snprintf(buf, buf_len,
"Main program %s is an image file", fip->name
);
return ERR_NO_SIGNATURE;
}
if (!fip->executable && !wup->project->anonymous_platform) {
snprintf(buf, buf_len,
"Main program %s is not executable", fip->name
);
return ERR_NO_SIGNATURE;
}
}
retval = setup_file(fip, fref, file_path, true, false);
if (retval == ERR_IN_PROGRESS) {
return retval;
} else if (retval) {
safe_strcpy(buf, "Can't link app version file");
return retval;
}
}

// set up input, output files
//
for (i=0; i<wup->input_files.size(); i++) {
fref = wup->input_files[i];
fip = fref.file_info;
get_pathname(fref.file_info, file_path, sizeof(file_path));
retval = setup_file(fip, fref, file_path, true, true);
if (retval == ERR_IN_PROGRESS) {
return retval;
} else if (retval) {
strcpy(buf, "Can't link input file");
return retval;
}
}
for (i=0; i<result->output_files.size(); i++) {
fref = result->output_files[i];
if (must_copy_file(fref, true)) continue;
fip = fref.file_info;
get_pathname(fref.file_info, file_path, sizeof(file_path));
retval = setup_file(fip, fref, file_path, false, true);
if (retval) {
strcpy(buf, "Can't link output file");
return retval;
}
}

link_user_files();
// don't check retval here

FILE *f = fopen(path, "w");
fclose(f);
return 0;
}

// Start a task in a slot directory.
// This includes setting up soft links,
// passing preferences, and starting the process
// set up the slot dir if needed
//
// Current dir is top-level BOINC dir
//
Expand All @@ -534,20 +625,22 @@ static int get_priority(bool is_high_priority) {
// else
// ACTIVE_TASK::task_state is PROCESS_EXECUTING
//
// If "test" is set, we're doing the API test; just run "test_app".
//
int ACTIVE_TASK::start(bool test) {
int ACTIVE_TASK::start() {
char exec_name[256], file_path[MAXPATHLEN], buf[MAXPATHLEN], exec_path[MAXPATHLEN];
char cmdline[80000]; // 64KB plus some extra
unsigned int i;
FILE_REF fref;
FILE_INFO *fip;
int retval;
APP_INIT_DATA aid;
bool high_priority;
#ifdef _WIN32
bool success = false;
LPVOID environment_block=NULL;
#endif

// is an async copy in progress for this job?
//
if (async_copy) {
if (log_flags.task_debug) {
msg_printf(wup->project, MSG_INFO,
Expand All @@ -563,29 +656,25 @@ int ACTIVE_TASK::start(bool test) {
strcpy(slot_dir, "slots/app_test");
}

// run it at above idle priority if it
// set up the slot dir
//
retval = setup_slot_dir(buf, sizeof(buf));
if (retval == ERR_IN_PROGRESS) {
set_task_state(PROCESS_COPY_PENDING, "start");
return 0;
}
if (retval) goto error;

// run process at above idle priority if it
// - uses coprocs
// - uses less than one CPU
// - is a wrapper
//
bool high_priority = false;
high_priority = false;
if (app_version->rsc_type()) high_priority = true;
if (app_version->avg_ncpus < 1) high_priority = true;
if (app_version->is_wrapper) high_priority = true;

// make sure the task files exist
//
FILE_INFO* fip = 0;
retval = gstate.task_files_present(result, true, &fip);
if (retval) {
snprintf(
buf, sizeof(buf),
"Task file %s: %s",
fip->name, boincerror(retval)
);
goto error;
}

current_cpu_time = checkpoint_cpu_time;
elapsed_time = checkpoint_elapsed_time;
fraction_done = checkpoint_fraction_done;
Expand Down Expand Up @@ -613,84 +702,35 @@ int ACTIVE_TASK::start(bool test) {
init_app_init_data(aid);
retval = write_app_init_file(aid);
if (retval) {
snprintf(buf, sizeof(buf), "Can't write init file: %s", boincerror(retval));
snprintf(buf, sizeof(buf),
"Can't write init file: %s", boincerror(retval)
);
goto error;
}

// set up applications files
// get main prog filename and path
//
if (test) {
safe_strcpy(exec_name, "test_app");
safe_strcpy(exec_path, "test_app");
} else {
safe_strcpy(exec_name, "");
}
safe_strcpy(exec_name, "");
for (i=0; i<app_version->app_files.size(); i++) {
fref = app_version->app_files[i];
fip = fref.file_info;
get_pathname(fip, file_path, sizeof(file_path));
if (fref.main_program) {
if (is_image_file(fip->name)) {
snprintf(buf, sizeof(buf), "Main program %s is an image file", fip->name);
retval = ERR_NO_SIGNATURE;
goto error;
}
if (!fip->executable && !wup->project->anonymous_platform) {
snprintf(buf, sizeof(buf), "Main program %s is not executable", fip->name);
retval = ERR_NO_SIGNATURE;
goto error;
}
get_pathname(fip, file_path, sizeof(file_path));
safe_strcpy(exec_name, fip->name);
safe_strcpy(exec_path, file_path);
}
retval = setup_file(fip, fref, file_path, true, false);
if (retval == ERR_IN_PROGRESS) {
set_task_state(PROCESS_COPY_PENDING, "start");
return 0;
} else if (retval) {
safe_strcpy(buf, "Can't link app version file");
goto error;
}
}
if (!strlen(exec_name)) {
safe_strcpy(buf, "No main program specified");
retval = ERR_NOT_FOUND;
goto error;
}

// set up input, output files
//
for (i=0; i<wup->input_files.size(); i++) {
fref = wup->input_files[i];
fip = fref.file_info;
get_pathname(fref.file_info, file_path, sizeof(file_path));
retval = setup_file(fip, fref, file_path, true, true);
if (retval == ERR_IN_PROGRESS) {
set_task_state(PROCESS_COPY_PENDING, "start");
return 0;
} else if (retval) {
safe_strcpy(buf, "Can't link input file");
goto error;
}
}
for (i=0; i<result->output_files.size(); i++) {
fref = result->output_files[i];
if (must_copy_file(fref, true)) continue;
fip = fref.file_info;
get_pathname(fref.file_info, file_path, sizeof(file_path));
retval = setup_file(fip, fref, file_path, false, true);
if (retval) {
safe_strcpy(buf, "Can't link output file");
goto error;
}
}

link_user_files();
// don't check retval here

// remove temporary exit file from last run
//
snprintf(file_path, sizeof(file_path), "%s/%s", slot_dir, TEMPORARY_EXIT_FILE);
snprintf(file_path, sizeof(file_path), "%s/%s",
slot_dir, TEMPORARY_EXIT_FILE
);
delete_project_owned_file(file_path, true);

if (cc_config.exit_before_start) {
Expand Down Expand Up @@ -1096,11 +1136,7 @@ int ACTIVE_TASK::start(bool test) {
// If using account-based sandboxing, use a helper app
// to do this, to set the right user ID
//
if (test) {
strcpy(buf, exec_path);
} else {
snprintf(buf, sizeof(buf), "../../%.1024s", exec_path);
}
snprintf(buf, sizeof(buf), "../../%.1024s", exec_path);
if (g_use_sandbox) {
char switcher_path[MAXPATHLEN];
snprintf(switcher_path, sizeof(switcher_path),
Expand Down Expand Up @@ -1149,9 +1185,6 @@ int ACTIVE_TASK::start(bool test) {
error:
// here on error; "buf" contains error message, "retval" is nonzero
//
if (test) {
return retval;
}

// if failed to run program, it's possible that the executable was munged.
// Verify the app version files to detect this
Expand All @@ -1176,6 +1209,7 @@ int ACTIVE_TASK::start(bool test) {
}

// Resume the task if it was previously running; otherwise start it
// "first_time" is set if the slot dir is empty
// Postcondition: "state" is set correctly
//
int ACTIVE_TASK::resume_or_start(bool first_time) {
Expand Down Expand Up @@ -1234,80 +1268,3 @@ int ACTIVE_TASK::resume_or_start(bool first_time) {
}
return 0;
}

// The following runs "test_app" and sends it various messages.
// Used for testing the runtime system.
//
void run_test_app() {
WORKUNIT wu;
PROJECT project;
APP app;
APP_VERSION av;
ACTIVE_TASK at;
ACTIVE_TASK_SET ats;
RESULT result;
int retval;

gstate.run_test_app = true;

wu.project = &project;
wu.app = &app;
wu.command_line = string("--critical_section");

safe_strcpy(app.name, "test app");
av.init();
av.avg_ncpus = 1;

safe_strcpy(result.name, "test result");
result.avp = &av;
result.wup = &wu;
result.project = &project;
result.app = &app;

at.result = &result;
at.wup = &wu;
at.app_version = &av;
at.max_elapsed_time = 1e6;
at.max_disk_usage = 1e14;
at.max_mem_usage = 1e14;
safe_strcpy(at.slot_dir, ".");

#if 0
// test file copy
//
ASYNC_COPY* ac = new ASYNC_COPY;
FILE_INFO fi;
retval = ac->init(&at, &fi, "big_file", "./big_file_copy");
if (retval) {
exit(1);
}
while (1) {
do_async_file_op();
if (at.async_copy == NULL) {
break;
}
}
fprintf(stderr, "done\n");
exit(0);
#endif
ats.active_tasks.push_back(&at);

unlink("boinc_finish_called");
unlink("boinc_lockfile");
unlink("boinc_temporary_exit");
unlink("stderr.txt");
retval = at.start(true);
if (retval) {
fprintf(stderr, "start() failed: %s\n", boincerror(retval));
}
while (1) {
gstate.now = dtime();
at.preempt(REMOVE_NEVER);
ats.poll();
boinc_sleep(.1);
at.unsuspend();
ats.poll();
boinc_sleep(.2);
//at.request_reread_prefs();
}
}
Loading

0 comments on commit 54266b2

Please sign in to comment.