Handle Unkown Tasks and RPC Error way better

This commit is contained in:
Samuel Lorch 2024-07-06 15:54:49 +02:00
parent 7574628a1d
commit 78ab8c9daf
4 changed files with 33 additions and 4 deletions

View file

@ -2,6 +2,8 @@ package server
import (
"context"
"errors"
"fmt"
"log/slog"
"net/http"
"strings"
@ -207,7 +209,7 @@ func updateWorkerTaskStatus(ctx context.Context) {
if strings.HasSuffix(err.Error(), constants.ErrTaskDoesNotExist.Error()) {
// Worker says it does not know of this task, mark it failed so that we don't asks the worker about it again and again
slog.ErrorContext(ctx, "Task is unknown by worker, Failing...", "err", err, "id", taskID)
_, err = db.Exec(ctx, "UPDATE tasks SET status = $2, log = log || $3 WHERE id = $1", taskID, constants.TASK_STATUS_FAILED, []string{"Task Failed because it is unknown to Assigned Worker"})
_, err = db.Exec(ctx, "UPDATE tasks SET status = $2, log = log || $3 WHERE id = $1", taskID, constants.TASK_STATUS_FAILED, []string{fmt.Sprintf("%v MASTER: Task Failed because it is unknown to Assigned Worker", time.Now())})
if err != nil {
slog.ErrorContext(ctx, "Updating Failed Task Status", "err", err)
return
@ -215,10 +217,17 @@ func updateWorkerTaskStatus(ctx context.Context) {
slog.Info("Updating task done", "id", taskID, "status", constants.TASK_STATUS_FAILED)
return
} else if errors.Is(err, constants.ErrRPCRequestTimeout) {
// We really don't know whats going on, might be slow response, oom, disk full or a bug
_, err = db.Exec(ctx, "UPDATE tasks SET status = $2, log = log || $3 WHERE id = $1", taskID, constants.TASK_STATUS_UNKNOWN, []string{fmt.Sprintf("%v MASTER: Status RPC Call Timed Out!", time.Now())})
if err != nil {
slog.ErrorContext(ctx, "Updating Unknown Task Status due to Timeout", "err", err)
return
}
} else {
slog.ErrorContext(ctx, "Getting Task Status", "err", err)
_, err = db.Exec(ctx, "UPDATE tasks SET status = $2 WHERE id = $1", taskID, constants.TASK_STATUS_UNKNOWN)
_, err = db.Exec(ctx, "UPDATE tasks SET status = $2, log = log || $3 WHERE id = $1", taskID, constants.TASK_STATUS_UNKNOWN, []string{fmt.Sprintf("%v MASTER: Status RPC Call Error: %v", time.Now(), err.Error())})
if err != nil {
slog.ErrorContext(ctx, "Updating Unknown Task Status", "err", err)
return