… | |
… | |
3404 | cf::cleanup "SIG$signal"; |
3404 | cf::cleanup "SIG$signal"; |
3405 | }; |
3405 | }; |
3406 | } |
3406 | } |
3407 | } |
3407 | } |
3408 | |
3408 | |
3409 | sub write_runtime { |
3409 | sub write_runtime_sync { |
3410 | my $runtime = "$LOCALDIR/runtime"; |
3410 | my $runtime = "$LOCALDIR/runtime"; |
3411 | |
3411 | |
3412 | # first touch the runtime file to show we are still running: |
3412 | # first touch the runtime file to show we are still running: |
3413 | # the fsync below can take a very very long time. |
3413 | # the fsync below can take a very very long time. |
3414 | |
3414 | |
… | |
… | |
3440 | and return; |
3440 | and return; |
3441 | |
3441 | |
3442 | warn "runtime file written.\n"; |
3442 | warn "runtime file written.\n"; |
3443 | |
3443 | |
3444 | 1 |
3444 | 1 |
|
|
3445 | } |
|
|
3446 | |
|
|
3447 | our $uuid_lock; |
|
|
3448 | our $uuid_skip; |
|
|
3449 | |
|
|
3450 | sub write_uuid_sync($) { |
|
|
3451 | $uuid_skip ||= $_[0]; |
|
|
3452 | |
|
|
3453 | return if $uuid_lock; |
|
|
3454 | local $uuid_lock = 1; |
|
|
3455 | |
|
|
3456 | my $uuid = "$LOCALDIR/uuid"; |
|
|
3457 | |
|
|
3458 | my $fh = aio_open "$uuid~", O_WRONLY | O_CREAT, 0644 |
|
|
3459 | or return; |
|
|
3460 | |
|
|
3461 | my $value = uuid_str $uuid_skip + uuid_seq uuid_cur; |
|
|
3462 | $uuid_skip = 0; |
|
|
3463 | |
|
|
3464 | (aio_write $fh, 0, (length $value), $value, 0) <= 0 |
|
|
3465 | and return; |
|
|
3466 | |
|
|
3467 | # always fsync - this file is important |
|
|
3468 | aio_fsync $fh |
|
|
3469 | and return; |
|
|
3470 | |
|
|
3471 | close $fh |
|
|
3472 | or return; |
|
|
3473 | |
|
|
3474 | aio_rename "$uuid~", $uuid |
|
|
3475 | and return; |
|
|
3476 | |
|
|
3477 | warn "uuid file written ($value).\n"; |
|
|
3478 | |
|
|
3479 | 1 |
|
|
3480 | |
|
|
3481 | } |
|
|
3482 | |
|
|
3483 | sub write_uuid($$) { |
|
|
3484 | my ($skip, $sync) = @_; |
|
|
3485 | |
|
|
3486 | $sync ? write_uuid_sync $skip |
|
|
3487 | : async { write_uuid_sync $skip }; |
3445 | } |
3488 | } |
3446 | |
3489 | |
3447 | sub emergency_save() { |
3490 | sub emergency_save() { |
3448 | my $freeze_guard = cf::freeze_mainloop; |
3491 | my $freeze_guard = cf::freeze_mainloop; |
3449 | |
3492 | |
… | |
… | |
3471 | warn "end emergency map save\n"; |
3514 | warn "end emergency map save\n"; |
3472 | |
3515 | |
3473 | warn "begin emergency database checkpoint\n"; |
3516 | warn "begin emergency database checkpoint\n"; |
3474 | BDB::db_env_txn_checkpoint $DB_ENV; |
3517 | BDB::db_env_txn_checkpoint $DB_ENV; |
3475 | warn "end emergency database checkpoint\n"; |
3518 | warn "end emergency database checkpoint\n"; |
|
|
3519 | |
|
|
3520 | warn "begin write uuid\n"; |
|
|
3521 | write_uuid_sync 1; |
|
|
3522 | warn "end write uuid\n"; |
3476 | }; |
3523 | }; |
3477 | |
3524 | |
3478 | warn "leave emergency perl save\n"; |
3525 | warn "leave emergency perl save\n"; |
3479 | } |
3526 | } |
3480 | |
3527 | |
… | |
… | |
3495 | warn "reloading..."; |
3542 | warn "reloading..."; |
3496 | |
3543 | |
3497 | warn "entering sync_job"; |
3544 | warn "entering sync_job"; |
3498 | |
3545 | |
3499 | cf::sync_job { |
3546 | cf::sync_job { |
3500 | cf::write_runtime; # external watchdog should not bark |
3547 | cf::write_runtime_sync; # external watchdog should not bark |
3501 | cf::emergency_save; |
3548 | cf::emergency_save; |
3502 | cf::write_runtime; # external watchdog should not bark |
3549 | cf::write_runtime_sync; # external watchdog should not bark |
3503 | |
3550 | |
3504 | warn "syncing database to disk"; |
3551 | warn "syncing database to disk"; |
3505 | BDB::db_env_txn_checkpoint $DB_ENV; |
3552 | BDB::db_env_txn_checkpoint $DB_ENV; |
3506 | |
3553 | |
3507 | # if anything goes wrong in here, we should simply crash as we already saved |
3554 | # if anything goes wrong in here, we should simply crash as we already saved |
… | |
… | |
3662 | |
3709 | |
3663 | if ($NOW >= $NEXT_RUNTIME_WRITE) { |
3710 | if ($NOW >= $NEXT_RUNTIME_WRITE) { |
3664 | $NEXT_RUNTIME_WRITE = List::Util::max $NEXT_RUNTIME_WRITE + 10, $NOW + 5.; |
3711 | $NEXT_RUNTIME_WRITE = List::Util::max $NEXT_RUNTIME_WRITE + 10, $NOW + 5.; |
3665 | Coro::async_pool { |
3712 | Coro::async_pool { |
3666 | $Coro::current->{desc} = "runtime saver"; |
3713 | $Coro::current->{desc} = "runtime saver"; |
3667 | write_runtime |
3714 | write_runtime_sync |
3668 | or warn "ERROR: unable to write runtime file: $!"; |
3715 | or warn "ERROR: unable to write runtime file: $!"; |
3669 | }; |
3716 | }; |
3670 | } |
3717 | } |
3671 | |
3718 | |
3672 | if (my $sig = shift @WAIT_FOR_TICK_BEGIN) { |
3719 | if (my $sig = shift @WAIT_FOR_TICK_BEGIN) { |