@cryptotaxi247 / netdata-1 / commits / 788ae9a44

spawn server improvements (#18115)

* spawn server improvements * fixed message

Costa Tsaousis committed Jul 11, 2024 at 16:52 UTC 788ae9a44d059d2ed66239d248bd791a9a453a0d

8 files changed +162 -135

src/daemon/main.c

+5 -17

@@ -324,6 +324,8 @@ static bool service_wait_exit(SERVICE_TYPE service, usec_t timeout_ut) {

324 void web_client_cache_destroy(void);

325

326 void netdata_cleanup_and_exit(int ret, const char *action, const char *action_result, const char *action_data) {

327 + netdata_exit = 1;

328 +

329 watcher_shutdown_begin();

330

331 nd_log_limits_unlimited();

@@ -344,6 +346,9 @@ void netdata_cleanup_and_exit(int ret, const char *action, const char *action_re

346 (void) rename(agent_crash_file, agent_incomplete_shutdown_file);

347 watcher_step_complete(WATCHER_STEP_ID_CREATE_SHUTDOWN_FILE);

348

349 + netdata_main_spawn_server_cleanup();

350 + watcher_step_complete(WATCHER_STEP_ID_DESTROY_MAIN_SPAWN_SERVER);

351 +

352 #ifdef ENABLE_DBENGINE

353 if(dbengine_enabled) {

354 for (size_t tier = 0; tier < storage_tiers; tier++)

@@ -486,9 +491,6 @@ void netdata_cleanup_and_exit(int ret, const char *action, const char *action_re

491 #endif

492 watcher_step_complete(WATCHER_STEP_ID_FREE_OPENSSL_STRUCTURES);

493

489 - netdata_main_spawn_server_cleanup();

490 - watcher_step_complete(WATCHER_STEP_ID_DESTROY_MAIN_SPAWN_SERVER);

491 -

494 (void) unlink(agent_incomplete_shutdown_file);

495 watcher_step_complete(WATCHER_STEP_ID_REMOVE_INCOMPLETE_SHUTDOWN_FILE);

496

@@ -667,8 +669,6 @@ void cancel_main_threads() {

669 }

670 }

671

670 - netdata_exit = 1;

671 -

672 while(found && max > 0) {

673 max -= step;

674 netdata_log_info("Waiting %d threads to finish...", found);

@@ -2179,18 +2179,6 @@ int netdata_main(int argc, char **argv) {

2179 delta_startup_time("initialize registry");

2180 registry_init();

2181

2182 - // fork the spawn server

2183 - delta_startup_time("fork the spawn server");

2184 -

2185 - /*

2186 - * Libuv uv_spawn() uses SIGCHLD internally:

2187 - * https://github.com/libuv/libuv/blob/cc51217a317e96510fbb284721d5e6bc2af31e33/src/unix/process.c#L485

2188 - * and inadvertently replaces the netdata signal handler which was setup during initialization.

2189 - * Thusly, we must explicitly restore the signal handler for SIGCHLD.

2190 - * Warning: extreme care is needed when mixing and matching POSIX and libuv.

2191 - */

2192 - signals_restore_SIGCHLD();

2193 -

2182 // ------------------------------------------------------------------------

2183 // initialize rrd, registry, health, rrdpush, etc.

2184

src/daemon/signals.c

+6 -60

@@ -2,6 +2,12 @@

3 #include "common.h"

5 +/*

6 + * IMPORTANT: Libuv uv_spawn() uses SIGCHLD internally:

7 + * https://github.com/libuv/libuv/blob/cc51217a317e96510fbb284721d5e6bc2af31e33/src/unix/process.c#L485

8 + * Extreme care is needed when mixing and matching POSIX and libuv.

9 + */

10 +

11 typedef enum signal_action {

12 NETDATA_SIGNAL_END_OF_LIST,

13 NETDATA_SIGNAL_IGNORE,

@@ -9,7 +15,6 @@ typedef enum signal_action {

15 NETDATA_SIGNAL_REOPEN_LOGS,

16 NETDATA_SIGNAL_RELOAD_HEALTH,

17 NETDATA_SIGNAL_FATAL,

12 - NETDATA_SIGNAL_CHILD,

18 } SIGNAL_ACTION;

20 static struct {

@@ -25,7 +30,6 @@ static struct {

30 { SIGHUP, "SIGHUP", 0, NETDATA_SIGNAL_REOPEN_LOGS },

31 { SIGUSR2, "SIGUSR2", 0, NETDATA_SIGNAL_RELOAD_HEALTH },

32 { SIGBUS, "SIGBUS", 0, NETDATA_SIGNAL_FATAL },

28 - { SIGCHLD, "SIGCHLD", 0, NETDATA_SIGNAL_CHILD },

34 // terminator

35 { 0, "NONE", 0, NETDATA_SIGNAL_END_OF_LIST }

@@ -93,18 +97,6 @@ void signals_init(void) {

97 }

98 }

96 -void signals_restore_SIGCHLD(void)

97 -{

98 - struct sigaction sa;

99 -

100 - sa.sa_flags = 0;

101 - sigfillset(&sa.sa_mask);

102 - sa.sa_handler = signal_handler;

103 -

104 - if(sigaction(SIGCHLD, &sa, NULL) == -1)

105 - netdata_log_error("SIGNAL: Failed to change signal handler for: SIGCHLD");

106 -}

107 -

100 void signals_reset(void) {

101 struct sigaction sa;

102 sigemptyset(&sa.sa_mask);

@@ -118,48 +110,6 @@ void signals_reset(void) {

110 }

111 }

112

121 -static void sigchild_handle() {

122 - int status;

123 - pid_t pid;

124 -

125 - // Loop to check for exited child processes

126 - while ((pid = waitpid((pid_t)(-1), &status, WNOHANG)) != 0) {

127 - if(pid == -1)

128 - break;

129 -

130 - if(WIFEXITED(status)) {

131 - nd_log(NDLS_DAEMON, NDLP_INFO,

132 - "DAEMON: child with pid %d exited normally with exit code %d",

133 - pid, WEXITSTATUS(status));

134 - }

135 - else if(WIFSIGNALED(status)) {

136 - if(WCOREDUMP(status))

137 - nd_log(NDLS_DAEMON, NDLP_INFO,

138 - "DAEMON: child with pid %d coredump'd due to signal %d",

139 - pid, WTERMSIG(status));

140 - else

141 - nd_log(NDLS_DAEMON, NDLP_INFO,

142 - "DAEMON: child with pid %d killed by signal %d",

143 - pid, WTERMSIG(status));

144 - }

145 - else if(WIFSTOPPED(status)) {

146 - nd_log(NDLS_DAEMON, NDLP_INFO,

147 - "DAEMON: child with pid %d stopped due to signal %d",

148 - pid, WSTOPSIG(status));

149 - }

150 - else if(WIFCONTINUED(status)) {

151 - nd_log(NDLS_DAEMON, NDLP_INFO,

152 - "DAEMON: child with pid %d continued due to signal %d",

153 - pid, SIGCONT);

154 - }

155 - else {

156 - nd_log(NDLS_COLLECTORS, NDLP_INFO,

157 - "DAEMON: child with pid %d reports unhandled status",

158 - pid);

159 - }

160 - }

161 -}

162 -

113 void signals_handle(void) {

114 while(1) {

115

@@ -211,10 +161,6 @@ void signals_handle(void) {

161 fatal("SIGNAL: Received %s. netdata now exits.", name);

162 break;

163

214 - case NETDATA_SIGNAL_CHILD:

215 - sigchild_handle();

216 - break;

217 -

164 default:

165 netdata_log_info("SIGNAL: Received %s. No signal handler configured. Ignoring it.", name);

166 break;

src/daemon/signals.h

-1

@@ -6,7 +6,6 @@

6 void signals_init(void);

7 void signals_block(void);

8 void signals_unblock(void);

9 -void signals_restore_SIGCHLD(void);

9 void signals_reset(void);

10 void signals_handle(void) NORETURN;

src/daemon/watcher.c

+3 -2

@@ -65,6 +65,7 @@ void *watcher_main(void *arg)

65 usec_t shutdown_start_time = now_monotonic_usec();

67 watcher_wait_for_step(WATCHER_STEP_ID_CREATE_SHUTDOWN_FILE);

68 + watcher_wait_for_step(WATCHER_STEP_ID_DESTROY_MAIN_SPAWN_SERVER);

69 watcher_wait_for_step(WATCHER_STEP_ID_DBENGINE_EXIT_MODE);

70 watcher_wait_for_step(WATCHER_STEP_ID_CLOSE_WEBRTC_CONNECTIONS);

71 watcher_wait_for_step(WATCHER_STEP_ID_DISABLE_MAINTENANCE_NEW_QUERIES_NEW_WEB_REQUESTS_NEW_STREAMING_CONNECTIONS_AND_ACLK);

@@ -105,6 +106,8 @@ void watcher_thread_start() {

106

107 watcher_steps[WATCHER_STEP_ID_CREATE_SHUTDOWN_FILE].msg =

108 "create shutdown file";

109 + watcher_steps[WATCHER_STEP_ID_DESTROY_MAIN_SPAWN_SERVER].msg =

110 + "destroy main spawn server";

111 watcher_steps[WATCHER_STEP_ID_DBENGINE_EXIT_MODE].msg =

112 "dbengine exit mode";

113 watcher_steps[WATCHER_STEP_ID_CLOSE_WEBRTC_CONNECTIONS].msg =

@@ -151,8 +154,6 @@ void watcher_thread_start() {

154 "remove pid file";

155 watcher_steps[WATCHER_STEP_ID_FREE_OPENSSL_STRUCTURES].msg =

156 "free openssl structures";

154 - watcher_steps[WATCHER_STEP_ID_DESTROY_MAIN_SPAWN_SERVER].msg =

155 - "destroy main spawn server";

157 watcher_steps[WATCHER_STEP_ID_REMOVE_INCOMPLETE_SHUTDOWN_FILE].msg =

158 "remove incomplete shutdown file";

159

src/daemon/watcher.h

+1 -1

@@ -7,6 +7,7 @@

8 typedef enum {

9 WATCHER_STEP_ID_CREATE_SHUTDOWN_FILE = 0,

10 + WATCHER_STEP_ID_DESTROY_MAIN_SPAWN_SERVER,

11 WATCHER_STEP_ID_DBENGINE_EXIT_MODE,

12 WATCHER_STEP_ID_CLOSE_WEBRTC_CONNECTIONS,

13 WATCHER_STEP_ID_DISABLE_MAINTENANCE_NEW_QUERIES_NEW_WEB_REQUESTS_NEW_STREAMING_CONNECTIONS_AND_ACLK,

@@ -30,7 +31,6 @@ typedef enum {

31 WATCHER_STEP_ID_CLOSE_SQL_DATABASES,

32 WATCHER_STEP_ID_REMOVE_PID_FILE,

33 WATCHER_STEP_ID_FREE_OPENSSL_STRUCTURES,

33 - WATCHER_STEP_ID_DESTROY_MAIN_SPAWN_SERVER,

34 WATCHER_STEP_ID_REMOVE_INCOMPLETE_SHUTDOWN_FILE,

36 // Always keep this as the last enum value

src/libnetdata/spawn_server/spawn_popen.c

+9 -5

@@ -3,14 +3,14 @@

3 #include "spawn_popen.h"

5 SPAWN_SERVER *netdata_main_spawn_server = NULL;

6 +static SPINLOCK netdata_main_spawn_server_spinlock = NETDATA_SPINLOCK_INITIALIZER;

8 bool netdata_main_spawn_server_init(const char *name, int argc, const char **argv) {

9 if(netdata_main_spawn_server == NULL) {

9 - static SPINLOCK spinlock = NETDATA_SPINLOCK_INITIALIZER;

10 - spinlock_lock(&spinlock);

10 + spinlock_lock(&netdata_main_spawn_server_spinlock);

11 if(netdata_main_spawn_server == NULL)

12 netdata_main_spawn_server = spawn_server_create(SPAWN_SERVER_OPTION_EXEC, name, NULL, argc, argv);

13 - spinlock_unlock(&spinlock);

13 + spinlock_unlock(&netdata_main_spawn_server_spinlock);

14 }

16 return netdata_main_spawn_server != NULL;

@@ -18,8 +18,12 @@ bool netdata_main_spawn_server_init(const char *name, int argc, const char **arg

19 void netdata_main_spawn_server_cleanup(void) {

20 if(netdata_main_spawn_server) {

21 - spawn_server_destroy(netdata_main_spawn_server);

22 - netdata_main_spawn_server = NULL;

21 + spinlock_lock(&netdata_main_spawn_server_spinlock);

22 + if(netdata_main_spawn_server) {

23 + spawn_server_destroy(netdata_main_spawn_server);

24 + netdata_main_spawn_server = NULL;

25 + }

26 + spinlock_unlock(&netdata_main_spawn_server_spinlock);

27 }

28 }

src/libnetdata/spawn_server/spawn_server.c

+137 -49

@@ -471,7 +471,7 @@ static void spawn_server_run_child(SPAWN_SERVER *server, SPAWN_REQUEST *request)

471 // Encoding and decoding of spawn server request argv type of data

472

473 // Function to encode argv or envp

474 -static void* encode_argv(const char **argv, size_t *out_size) {

474 +static void* argv_encode(const char **argv, size_t *out_size) {

475 size_t buffer_size = 1024; // Initial buffer size

476 size_t buffer_used = 0;

477 char *buffer = mallocz(buffer_size);

@@ -505,7 +505,7 @@ static void* encode_argv(const char **argv, size_t *out_size) {

505 }

506

507 // Function to decode argv or envp

508 -static const char** decode_argv(const char *buffer, size_t size) {

508 +static const char** argv_decode(const char *buffer, size_t size) {

509 size_t count = 0;

510 const char *ptr = buffer;

511 while (ptr < buffer + size) {

@@ -529,6 +529,54 @@ static const char** decode_argv(const char *buffer, size_t size) {

529 return argv;

530 }

531

532 +static BUFFER *argv_to_cmdline_buffer(const char **argv) {

533 + BUFFER *wb = buffer_create(0, NULL);

534 +

535 + for(size_t i = 0; argv[i] ;i++) {

536 + const char *s = argv[i];

537 + size_t len = strlen(s);

538 + buffer_need_bytes(wb, len * 2 + 1);

539 +

540 + bool needs_quotes = false;

541 + for(const char *c = s; !needs_quotes && *c ; c++) {

542 + switch(*c) {

543 + case ' ':

544 + case '\v':

545 + case '\t':

546 + case '\n':

547 + case '"':

548 + needs_quotes = true;

549 + break;

550 +

551 + default:

552 + break;

553 + }

554 + }

555 +

556 + if(needs_quotes && buffer_strlen(wb))

557 + buffer_strcat(wb, " \"");

558 + else

559 + buffer_putc(wb, ' ');

560 +

561 + for(const char *c = s; *c ; c++) {

562 + switch(*c) {

563 + case '"':

564 + buffer_putc(wb, '\\');

565 + // fall through

566 +

567 + default:

568 + buffer_putc(wb, *c);

569 + break;

570 + }

571 + }

572 +

573 + if(needs_quotes)

574 + buffer_strcat(wb, "\"");

575 + }

576 +

577 + return wb;

578 +}

579 +

580 // --------------------------------------------------------------------------------------------------------------------

581 // Sending and receiving requests

582

@@ -600,12 +648,12 @@ static bool spawn_server_send_request(ND_UUID *magic, SPAWN_REQUEST *request) {

648 bool ret = false;

649

650 size_t env_size = 0;

603 - void *encoded_env = encode_argv(request->environment, &env_size);

651 + void *encoded_env = argv_encode(request->environment, &env_size);

652 if (!encoded_env)

653 goto cleanup;

654

655 size_t argv_size = 0;

608 - void *encoded_argv = encode_argv(request->argv, &argv_size);

656 + void *encoded_argv = argv_encode(request->argv, &argv_size);

657 if (!encoded_argv)

658 goto cleanup;

659

@@ -684,6 +732,19 @@ cleanup:

732 return ret;

733 }

734

735 +static void request_free(SPAWN_REQUEST *rq) {

736 + if(rq->fds[0] != -1) close(rq->fds[0]);

737 + if(rq->fds[1] != -1) close(rq->fds[1]);

738 + if(rq->fds[2] != -1) close(rq->fds[2]);

739 + if(rq->fds[3] != -1) close(rq->fds[3]);

740 + if(rq->socket != -1) close(rq->socket);

741 + freez((void *)rq->argv);

742 + freez((void *)rq->environment);

743 + freez((void *)rq->data);

744 + freez((void *)rq->cmdline);

745 + freez((void *)rq);

746 +}

747 +

748 static void spawn_server_receive_request(int sock, SPAWN_SERVER *server) {

749 struct msghdr msg = {0};

750 struct iovec iov[7];

@@ -695,7 +756,7 @@ static void spawn_server_receive_request(int sock, SPAWN_SERVER *server) {

756 ND_UUID magic = UUID_ZERO;

757 SPAWN_INSTANCE_TYPE type;

758 char cmsgbuf[CMSG_SPACE(sizeof(int) * SPAWN_SERVER_TRANSFER_FDS)];

698 - char *envp = NULL, *argv = NULL, *data = NULL;

759 + char *envp_encoded = NULL, *argv_encoded = NULL, *data = NULL;

760 int stdin_fd = -1, stdout_fd = -1, stderr_fd = -1, custom_fd = -1;

761

762 // First recvmsg() to read sizes and control message

@@ -791,14 +852,13 @@ static void spawn_server_receive_request(int sock, SPAWN_SERVER *server) {

852 nd_log(NDLS_COLLECTORS, NDLP_ERR,

853 "SPAWN SERVER: invalid file descriptors received, stdin = %d, stdout = %d, stderr = %d",

854 stdin_fd, stdout_fd, stderr_fd);

794 - close(sock);

855 goto cleanup;

856 }

857

858 // Second recvmsg() to read buffer contents

799 - iov[0].iov_base = envp = mallocz(env_size);

859 + iov[0].iov_base = envp_encoded = mallocz(env_size);

860 iov[0].iov_len = env_size;

801 - iov[1].iov_base = argv = mallocz(argv_size);

861 + iov[1].iov_base = argv_encoded = mallocz(argv_size);

862 iov[1].iov_len = argv_size;

863 iov[2].iov_base = data = mallocz(data_size);

864 iov[2].iov_len = data_size;

@@ -811,15 +871,14 @@ static void spawn_server_receive_request(int sock, SPAWN_SERVER *server) {

871 ssize_t total_bytes_received = recvmsg(sock, &msg, 0);

872 if (total_bytes_received < 0) {

873 nd_log(NDLS_COLLECTORS, NDLP_ERR, "SPAWN SERVER: failed to recvmsg() the second part of the request.");

814 - close(sock);

874 goto cleanup;

875 }

876

877 // fprintf(stderr, "SPAWN SERVER: received request %zu (fds: %d, %d, %d, %d)\n", request_id,

878 // stdin_fd, stdout_fd, stderr_fd, custom_fd);

879

821 - SPAWN_REQUEST *request = mallocz(sizeof(*request));

822 - *request = (SPAWN_REQUEST){

880 + SPAWN_REQUEST *rq = mallocz(sizeof(*rq));

881 + *rq = (SPAWN_REQUEST){

882 .pid = 0,

883 .request_id = request_id,

884 .socket = sock,

@@ -829,50 +888,78 @@ static void spawn_server_receive_request(int sock, SPAWN_SERVER *server) {

888 [2] = stderr_fd,

889 [3] = custom_fd,

890 },

832 - .environment = decode_argv(envp, env_size),

833 - .argv = decode_argv(argv, argv_size),

891 + .environment = argv_decode(envp_encoded, env_size),

892 + .argv = argv_decode(argv_encoded, argv_size),

893 .data = data,

894 .data_size = data_size,

895 .type = type

896 };

897

898 + switch(rq->type) {

899 + case SPAWN_INSTANCE_TYPE_EXEC:

900 + if(rq->argv) {

901 + CLEAN_BUFFER *wb = argv_to_cmdline_buffer(rq->argv);

902 + rq->cmdline = strdupz(buffer_tostring(wb));

903 + }

904 + break;

905 +

906 + case SPAWN_INSTANCE_TYPE_CALLBACK:

907 + rq->cmdline = strdupz("callback() function");

908 + break;

909 +

910 + default:

911 + rq->cmdline = strdupz("[unknown request type]");

912 + break;

913 + }

914 +

915 pid_t pid = fork();

916 if (pid == 0) {

917 // the child

842 - spawn_server_run_child(server, request);

918 + spawn_server_run_child(server, rq);

919 exit(63);

920 }

921 else if (pid > 0) {

922 // the parent

847 - request->environment = NULL; // will be free'd at cleanup

848 - request->argv = NULL; // will be free'd at cleanup

849 - request->data = NULL; // will be free'd at cleanup

850 - request->data_size = 0;

851 - request->pid = pid;

852 - request->fds[0] = -1;

853 - request->fds[1] = -1;

854 - request->fds[2] = -1;

855 - request->fds[3] = -1;

856 - DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(spawn_server_requests, request, prev, next);

923 + rq->pid = pid;

924 +

925 + // do not keep data we don't need at the parent

926 + freez(envp_encoded); freez((void *)rq->environment); rq->environment = NULL;

927 + freez(argv_encoded); freez((void *)rq->argv); rq->argv = NULL;

928 + freez((void *)rq->data); rq->data = NULL;

929 + rq->data_size = 0;

930 +

931 + // do not keep fds we don't need at the parent

932 + if(rq->fds[0] != -1) { close(rq->fds[0]); rq->fds[0] = -1; }

933 + if(rq->fds[1] != -1) { close(rq->fds[1]); rq->fds[1] = -1; }

934 + if(rq->fds[2] != -1) { close(rq->fds[2]); rq->fds[2] = -1; }

935 + if(rq->fds[3] != -1) { close(rq->fds[3]); rq->fds[3] = -1; }

936 +

937 + // keep it in the list

938 + DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(spawn_server_requests, rq, prev, next);

939

940 // do not fork this socket on other children

859 - sock_setcloexec(request->socket);

860 - }

861 - else {

862 - nd_log(NDLS_COLLECTORS, NDLP_ERR, "SPAWN SERVER: Failed to fork() child.");

863 - spawn_server_send_status_failure(stdout_fd);

864 - // the other allocations (envp, argv, data) will be free'd at cleanup

865 - freez((void *)request);

866 - close(sock);

941 + sock_setcloexec(rq->socket);

942 +

943 + // done with this

944 + return;

945 }

946

947 + nd_log(NDLS_COLLECTORS, NDLP_ERR, "SPAWN SERVER: Failed to fork() child.");

948 + spawn_server_send_status_failure(stdout_fd);

949 + // the other allocations (envp, argv, data) will be free'd at cleanup

950 + request_free(rq);

951 + freez(envp_encoded);

952 + freez(argv_encoded);

953 + return;

954 +

955 cleanup:

956 + close(sock);

957 if(stdin_fd != -1) close(stdin_fd);

958 if(stdout_fd != -1) close(stdout_fd);

959 if(stderr_fd != -1) close(stderr_fd);

960 if(custom_fd != -1) close(custom_fd);

874 - freez(envp);

875 - freez(argv);

961 + freez(envp_encoded);

962 + freez(argv_encoded);

963 freez(data);

964 }

965

@@ -906,6 +993,8 @@ static void spawn_server_process_sigchld(void) {

993 if(pid == -1)

994 break;

995

996 + errno_clear();

997 +

998 SPAWN_REQUEST *rq = find_request_by_pid(pid);

999 size_t request_id = rq ? rq->request_id : 0;

1000 bool send_report_remove_request = false;

@@ -913,45 +1002,44 @@ static void spawn_server_process_sigchld(void) {

1002 if(WIFEXITED(status)) {

1003 if(WEXITSTATUS(status))

1004 nd_log(NDLS_COLLECTORS, NDLP_INFO,

916 - "SPAWN SERVER: child with pid %d (request %zu) exited normally with exit code %d",

917 - pid, request_id, WEXITSTATUS(status));

1005 + "SPAWN SERVER: child with pid %d (request %zu) exited with exit code %d: %s",

1006 + pid, request_id, WEXITSTATUS(status), rq ? rq->cmdline : "[request not found]");

1007 send_report_remove_request = true;

1008 }

1009 else if(WIFSIGNALED(status)) {

1010 if(WCOREDUMP(status))

1011 nd_log(NDLS_COLLECTORS, NDLP_INFO,

923 - "SPAWN SERVER: child with pid %d (request %zu) coredump'd due to signal %d",

924 - pid, request_id, WTERMSIG(status));

1012 + "SPAWN SERVER: child with pid %d (request %zu) coredump'd due to signal %d: %s",

1013 + pid, request_id, WTERMSIG(status), rq ? rq->cmdline : "[request not found]");

1014 else

1015 nd_log(NDLS_COLLECTORS, NDLP_INFO,

927 - "SPAWN SERVER: child with pid %d (request %zu) killed by signal %d",

928 - pid, request_id, WTERMSIG(status));

1016 + "SPAWN SERVER: child with pid %d (request %zu) killed by signal %d: %s",

1017 + pid, request_id, WTERMSIG(status), rq ? rq->cmdline : "[request not found]");

1018 send_report_remove_request = true;

1019 }

1020 else if(WIFSTOPPED(status)) {

1021 nd_log(NDLS_COLLECTORS, NDLP_INFO,

933 - "SPAWN SERVER: child with pid %d (request %zu) stopped due to signal %d",

934 - pid, request_id, WSTOPSIG(status));

1022 + "SPAWN SERVER: child with pid %d (request %zu) stopped due to signal %d: %s",

1023 + pid, request_id, WSTOPSIG(status), rq ? rq->cmdline : "[request not found]");

1024 send_report_remove_request = false;

1025 }

1026 else if(WIFCONTINUED(status)) {

1027 nd_log(NDLS_COLLECTORS, NDLP_INFO,

939 - "SPAWN SERVER: child with pid %d (request %zu) continued due to signal %d",

940 - pid, request_id, SIGCONT);

1028 + "SPAWN SERVER: child with pid %d (request %zu) continued due to signal %d: %s",

1029 + pid, request_id, SIGCONT, rq ? rq->cmdline : "[request not found]");

1030 send_report_remove_request = false;

1031 }

1032 else {

1033 nd_log(NDLS_COLLECTORS, NDLP_INFO,

945 - "SPAWN SERVER: child with pid %d (request %zu) reports unhandled status",

946 - pid, request_id);

1034 + "SPAWN SERVER: child with pid %d (request %zu) reports unhandled status: %s",

1035 + pid, request_id, rq ? rq->cmdline : "[request not found]");

1036 send_report_remove_request = false;

1037 }

1038

1039 if(send_report_remove_request && rq) {

1040 spawn_server_send_status_exit(rq->socket, status);

952 - close(rq->socket);

1041 DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(spawn_server_requests, rq, prev, next);

954 - freez(rq);

1042 + request_free(rq);

1043 }

1044 }

1045 }

@@ -961,7 +1049,7 @@ static void signals_unblock(void) {

1049 sigfillset(&sigset);

1050

1051 if(pthread_sigmask(SIG_UNBLOCK, &sigset, NULL) == -1) {

964 - netdata_log_error("SIGNAL: Could not unblock signals for threads");

1052 + netdata_log_error("SPAWN SERVER: Could not unblock signals for threads");

1053 }

1054 }

1055

src/libnetdata/spawn_server/spawn_server.h

@@ -22,6 +22,7 @@ typedef enum __attribute__((packed)) {

22 // this is only used publicly for SPAWN_INSTANCE_TYPE_CALLBACK

23 // which is not available in Windows

24 typedef struct spawn_request {

25 + const char *cmdline;

26 size_t request_id;

27 pid_t pid;

28 int socket;