Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

validator: handle transient errors #5803

Merged
merged 1 commit into from
Sep 10, 2024
Merged

validator: handle transient errors #5803

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion sched/db_purge.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -590,7 +590,7 @@ int purge_and_archive_results(DB_WORKUNIT& wu, int& number_results) {
retval = result.delete_from_db();
if (retval) {
log_messages.printf(MSG_CRITICAL,
"Couldn't delete result [%d] from database\n", result.id
"Couldn't delete result [%lu] from database\n", result.id
);
return retval;
}
Expand Down
34 changes: 26 additions & 8 deletions sched/script_validator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,6 +104,8 @@ void validate_handler_usage() {
);
}

// see validate_util2.h for return values
//
int init_result(RESULT& result, void*&) {
if (init_script.empty()) {
return 0;
Expand All @@ -120,7 +122,6 @@ int init_result(RESULT& result, void*&) {
return retval;
}


char cmd[4096];
sprintf(cmd, "../bin/%s", init_script[0].c_str());
for (i=1; i<init_script.size(); i++) {
Expand All @@ -139,10 +140,19 @@ int init_result(RESULT& result, void*&) {
}
}
retval = system(cmd);
if (retval) {
return retval;
if (WIFEXITED(retval)) {
int s = WEXITSTATUS(retval);
if (!s) return 0;
if (s == VAL_RESULT_TRANSIENT_ERROR) {
return VAL_RESULT_TRANSIENT_ERROR;
}
log_messages.printf(MSG_CRITICAL,
"init script %s failed: %d\n", cmd, s
);
return -1;
}
return 0;
log_messages.printf(MSG_CRITICAL, "init script %s didn't exit\n", cmd);
return -1;
}

int compare_results(RESULT& r1, void*, RESULT const& r2, void*, bool& match) {
Expand Down Expand Up @@ -196,12 +206,20 @@ int compare_results(RESULT& r1, void*, RESULT const& r2, void*, bool& match) {
}
}
retval = system(cmd);
if (retval) {
if (WIFEXITED(retval)) {
int s = WEXITSTATUS(retval);
if (s == 0) {
match = true;
return 0;
}
if (s == VAL_RESULT_TRANSIENT_ERROR) {
return VAL_RESULT_TRANSIENT_ERROR;
}
match = false;
} else {
match = true;
return 0;
}
return 0;
log_messages.printf(MSG_CRITICAL, "compare script %s didn't exit\n", cmd);
return -1;
}

int cleanup_result(RESULT const&, void*) {
Expand Down
81 changes: 63 additions & 18 deletions sched/validate_util2.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,7 +65,7 @@ int check_set(
) {
vector<void*> data;
vector<bool> had_error;
int i, j, neq = 0, n, retval;
int i, j, neq = 0, n, retval=0;
int min_valid = wu.min_quorum/2+1;

retry = false;
Expand All @@ -83,29 +83,35 @@ int check_set(
int suspicious_results = 0;
for (i=0; i<n; i++) {
retval = init_result(results[i], data[i]);
if (retval == ERR_OPENDIR) {
switch (retval) {
case 0:
good_results++;
break;
case ERR_OPENDIR:
case VAL_RESULT_TRANSIENT_ERROR:
log_messages.printf(MSG_CRITICAL,
"check_set: init_result([RESULT#%lu %s]) transient failure\n",
results[i].id, results[i].name
);
retry = true;
had_error[i] = true;
} else if (retval == VAL_RESULT_SUSPICIOUS) {
break;
case VAL_RESULT_SUSPICIOUS:
log_messages.printf(MSG_NORMAL,
"[RESULT#%lu %s] considered to be suspicious\n",
results[i].id, results[i].name
);
suspicious_results++;
} else if (retval) {
break;
default:
log_messages.printf(MSG_CRITICAL,
"check_set: init_result([RESULT#%lu %s]) failed: %s\n",
results[i].id, results[i].name, boincerror(retval)
);
results[i].outcome = RESULT_OUTCOME_VALIDATE_ERROR;
results[i].validate_state = VALIDATE_STATE_INVALID;
had_error[i] = true;
} else {
good_results++;
break;
}
}

Expand Down Expand Up @@ -141,14 +147,29 @@ int check_set(
if (i == j) {
++neq;
matches[j] = true;
} else if (compare_results(results[i], data[i], results[j], data[j], match)) {
continue;
}
retval = compare_results(
results[i], data[i], results[j], data[j], match
);
switch (retval) {
case ERR_OPENDIR:
case VAL_RESULT_TRANSIENT_ERROR:
retry = true;
retval = 0;
goto cleanup;
case 0:
if (match) {
++neq;
matches[j] = true;
}
break;
default:
log_messages.printf(MSG_CRITICAL,
"generic_check_set: check_pair_with_data([RESULT#%lu %s], [RESULT#%lu %s]) failed\n",
"check_set(): compare_results([RESULT#%lu %s], [RESULT#%lu %s]) failed\n",
results[i].id, results[i].name, results[j].id, results[j].name
);
} else if (match) {
++neq;
matches[j] = true;
goto cleanup;
}
}
if (neq >= min_valid) {
Expand All @@ -164,14 +185,15 @@ int check_set(
}
}

retval = 0;
cleanup:

for (i=0; i<n; i++) {
cleanup_result(results[i], data[i]);
}
return 0;
return retval;
}

// a straggler instance has arrived after the WU is already validated.
// r1 is the new result; r2 is canonical result
//
void check_pair(RESULT& r1, RESULT& r2, bool& retry) {
Expand All @@ -182,14 +204,18 @@ void check_pair(RESULT& r1, RESULT& r2, bool& retry) {

retry = false;
retval = init_result(r1, data1);
if (retval == ERR_OPENDIR) {
switch (retval) {
case ERR_OPENDIR:
case VAL_RESULT_TRANSIENT_ERROR:
log_messages.printf(MSG_CRITICAL,
"check_pair: init_result([RESULT#%lu %s]) transient failure 1\n",
r1.id, r1.name
);
retry = true;
return;
} else if (retval) {
case 0:
break;
default:
log_messages.printf(MSG_CRITICAL,
"check_pair: init_result([RESULT#%lu %s]) perm failure 1\n",
r1.id, r1.name
Expand All @@ -200,15 +226,19 @@ void check_pair(RESULT& r1, RESULT& r2, bool& retry) {
}

retval = init_result(r2, data2);
if (retval == ERR_OPENDIR) {
switch (retval) {
case ERR_OPENDIR:
case VAL_RESULT_TRANSIENT_ERROR:
log_messages.printf(MSG_CRITICAL,
"check_pair: init_result([RESULT#%lu %s]) transient failure 2\n",
r2.id, r2.name
);
cleanup_result(r1, data1);
retry = true;
return;
} else if (retval) {
case 0:
break;
default:
log_messages.printf(MSG_CRITICAL,
"check_pair: init_result([RESULT#%lu %s]) perm failure2\n",
r2.id, r2.name
Expand All @@ -220,7 +250,22 @@ void check_pair(RESULT& r1, RESULT& r2, bool& retry) {
}

retval = compare_results(r1, data1, r2, data2, match);
r1.validate_state = match?VALIDATE_STATE_VALID:VALIDATE_STATE_INVALID;
switch (retval) {
case ERR_OPENDIR:
case VAL_RESULT_TRANSIENT_ERROR:
retry = true;
break;
case 0:
r1.validate_state = match?VALIDATE_STATE_VALID:VALIDATE_STATE_INVALID;
break;
default:
log_messages.printf(MSG_CRITICAL,
"check_pair: compare_results([RESULT#%lu RESULT#%lu]) perm failure\n",
r1.id, r2.id
);
r1.validate_state = VALIDATE_STATE_INVALID;
break;
}
cleanup_result(r1, data1);
cleanup_result(r2, data2);
}
14 changes: 11 additions & 3 deletions sched/validate_util2.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,12 +22,20 @@

#include "boinc_db_types.h"

// special return values of init_result():
// return values of init_result() and compare_result()
// (in addition to usual error codes)
//
#define VAL_RESULT_SUSPICIOUS 1
// if an "adaptive replication" result looks suspicious
// the result looks 'suspicious'.
// if we're using adaptive replication and this is the only copy,
// create a 2nd replica rather than accepting it.
// (Are any projects using this?)
#define VAL_RESULT_LONG_TERM_FAIL 2
// host is unlikely to handle this app version; stop using
// host is unlikely to handle this app version; stop using it
#define VAL_RESULT_TRANSIENT_ERROR 3
// a transient error happened (e.g. couldn't read file due to NFS failure).
// Retry validation later.
// ERR_OPENDIR is also treated this way.

extern int init_result(RESULT&, void*&);
extern int compare_results(RESULT &, void*, RESULT const&, void*, bool&);
Expand Down
Loading