@cryptotaxi247 / netdata
Code Commits Issues Pulls Sessions CI/CD
master
master
h 188 lines 7.09 KB
Raw
1 // SPDX-License-Identifier: GPL-3.0-or-later
2
3 #ifndef NETDATA_STATUS_FILE_H
4 #define NETDATA_STATUS_FILE_H
5
6 #include "libnetdata/libnetdata.h"
7 #include "daemon/config/netdata-conf-profile.h"
8 #include "database/rrd-database-mode.h"
9 #include "database/rrd-metadata.h"
10 #include "claim/cloud-status.h"
11 #include "machine-guid.h"
12 #include "status-file-dmi.h"
13
14 #define STATUS_FILE_VERSION 28
15
16 typedef enum {
17 DAEMON_STATUS_NONE,
18 DAEMON_STATUS_INITIALIZING,
19 DAEMON_STATUS_RUNNING,
20 DAEMON_STATUS_EXITING,
21 DAEMON_STATUS_EXITED,
22 } DAEMON_STATUS;
23 ENUM_STR_DEFINE_FUNCTIONS_EXTERN(DAEMON_STATUS);
24
25 typedef enum {
26 DAEMON_OS_TYPE_UNKNOWN,
27 DAEMON_OS_TYPE_LINUX,
28 DAEMON_OS_TYPE_FREEBSD,
29 DAEMON_OS_TYPE_MACOS,
30 DAEMON_OS_TYPE_WINDOWS,
31 } DAEMON_OS_TYPE;
32 ENUM_STR_DEFINE_FUNCTIONS_EXTERN(DAEMON_OS_TYPE);
33
34 typedef struct daemon_status_file {
35 uint32_t v; // the version of the status file
36
37 char version[32]; // the netdata version
38 DAEMON_STATUS status; // the daemon status
39 EXIT_REASON exit_reason; // the exit reason (maybe empty)
40 ND_PROFILE profile; // the profile of the agent
41 DAEMON_OS_TYPE os_type;
42 RRD_DB_MODE db_mode;
43 CLOUD_STATUS cloud_status;
44 uint8_t db_tiers;
45 bool kubernetes;
46 bool sentry_available; // true when sentry support is compiled in
47
48 time_t boottime; // system boottime
49 time_t uptime; // netdata uptime
50 usec_t timestamp_ut; // the timestamp of the status file
51 char timestamp_ut_rfc3339[RFC3339_MAX_LENGTH]; // pre-calculated RFC3339 for async-signal-safe use
52 size_t restarts; // the number of times this agent has restarted (ever)
53 size_t crashes; // the number of times this agent has crashed (ever)
54 size_t posts; // the number of posts to the backend
55 ssize_t reliability; // consecutive restarts: > 0 reliable, < 0 crashing
56 pid_t pid; // the process id of the netdata agent
57
58 ND_MACHINE_GUID host_id; // the machine guid of the system
59
60 ND_UUID boot_id; // the boot id of the system
61 ND_UUID invocation; // the netdata invocation id generated the file
62 ND_UUID node_id; // the Netdata Cloud node id of the agent
63 ND_UUID claim_id; // the Netdata Cloud claim id of the agent
64 ND_UUID machine_id; // the unique machine id of the system
65
66 struct {
67 usec_t init_started_ut;
68 time_t init;
69 usec_t exit_started_ut;
70 time_t exit;
71 } timings;
72
73 uint64_t oom_protection;
74 uint64_t netdata_max_rss;
75 OS_SYSTEM_MEMORY memory;
76 OS_SYSTEM_DISK_SPACE var_cache;
77
78 struct {
79 uint64_t dbengine; // Size of dbengine files
80 uint64_t sqlite; // Size of sqlite files
81 uint64_t other; // Size of other files (total - dbengine - sqlite)
82 usec_t last_updated_ut; // Last time the footprint was updated (microseconds)
83 char last_updated_ut_rfc3339[RFC3339_MAX_LENGTH]; // pre-calculated RFC3339 for async-signal-safe use
84 } disk_footprint;
85
86 // Metrics statistics
87 RRDSTATS_METADATA metrics_metadata;
88
89 char install_type[32];
90 char architecture[32]; // ECS: host.architecture
91 char virtualization[32];
92 char container[32];
93 char kernel_version[32]; // ECS: os.kernel
94 char os_name[32]; // ECS: os.name
95 char os_version[32]; // ECS: os.version
96 char os_id[64]; // ECS: os.family
97 char os_id_like[64]; // ECS: os.platform
98 char timezone[32];
99 char cloud_provider_type[32];
100 char cloud_instance_type[32];
101 char cloud_instance_region[32];
102 bool read_system_info;
103 size_t system_cpus;
104
105 DMI_INFO hw;
106
107 struct {
108 // normalized information from cloud provider and h/w information
109 char vendor[64];
110 char name[96];
111 char type[16];
112 } product;
113
114 char stack_traces[63]; // the backend for capturing stack traces
115
116 struct {
117 SPINLOCK spinlock;
118 long line;
119 char filename[256];
120 char function[128];
121 char errno_str[64];
122 char message[512];
123 char stack_trace[4096];
124 char thread[ND_THREAD_TAG_MAX + 1];
125 pid_t thread_id;
126 SIGNAL_CODE signal_code;
127 uintptr_t fault_address;
128 uint32_t worker_job_id;
129 bool sentry; // true when the error was also reported to sentry
130 } fatal;
131 } DAEMON_STATUS_FILE;
132
133 // these are used instead of locks when locks cannot be used (signal handler, out of memory, etc)
134 #define dsf_acquire(ds) __atomic_load_n(&(ds).v, __ATOMIC_ACQUIRE)
135 #define dsf_release(ds) __atomic_store_n(&(ds).v, (ds).v, __ATOMIC_RELEASE)
136
137 // saves the current status
138 void daemon_status_file_update_status(DAEMON_STATUS status);
139
140 // returns true when the event is duplicate and should not be reported again
141 bool daemon_status_file_deadly_signal_received(EXIT_REASON reason, SIGNAL_CODE code, void *fault_address, bool chained_handler);
142
143 // check for a crash
144 void daemon_status_file_check_crash(void);
145
146 bool daemon_status_file_has_last_crashed(DAEMON_STATUS_FILE *ds);
147 bool daemon_status_file_was_incomplete_shutdown(void);
148
149 void daemon_status_file_startup_step(const char *step);
150 void daemon_status_file_shutdown_step(const char *step, const char *step_timings);
151 void daemon_status_file_shutdown_timeout(BUFFER *trace);
152
153 void daemon_status_file_init(void);
154 void daemon_status_file_register_fatal(const char *filename, const char *function, const char *message, const char *errno_str, const char *stack_trace, long line);
155
156 const char *daemon_status_file_get_install_type(void);
157 const char *daemon_status_file_get_architecture(void);
158 const char *daemon_status_file_get_virtualization(void);
159 const char *daemon_status_file_get_container(void);
160 const char *daemon_status_file_get_os_name(void);
161 const char *daemon_status_file_get_os_version(void);
162 const char *daemon_status_file_get_os_id(void);
163 const char *daemon_status_file_get_os_id_like(void);
164 const char *daemon_status_file_get_timezone(void);
165 const char *daemon_status_file_get_cloud_provider_type(void);
166 const char *daemon_status_file_get_cloud_instance_type(void);
167 const char *daemon_status_file_get_cloud_instance_region(void);
168
169 const char *daemon_status_file_get_fatal_filename(void);
170 const char *daemon_status_file_get_fatal_function(void);
171 const char *daemon_status_file_get_fatal_message(void);
172 const char *daemon_status_file_get_fatal_errno(void);
173 const char *daemon_status_file_get_fatal_stack_trace(void);
174 const char *daemon_status_file_get_fatal_thread(void);
175 const char *daemon_status_file_get_stack_trace_backend(void);
176 const char *daemon_status_file_get_sys_vendor(void);
177 const char *daemon_status_file_get_product_name(void);
178 const char *daemon_status_file_get_product_type(void);
179
180 pid_t daemon_status_file_get_fatal_thread_id(void);
181 long daemon_status_file_get_fatal_line(void);
182 DAEMON_STATUS daemon_status_file_get_status(void);
183 size_t daemon_status_file_get_restarts(void);
184 ssize_t daemon_status_file_get_reliability(void);
185 ND_MACHINE_GUID daemon_status_file_get_host_id(void);
186 size_t daemon_status_file_get_fatal_worker_job_id(void);
187
188 #endif //NETDATA_STATUS_FILE_H