projects / lttng-tools.git / blame
| Commit | Line | Data |
|---|---|---|
| 2df048c2 DG |
1 | .TH LTTNG_HEALTH_CHECK 3 2012-09-19 "LTTng" "LTTng Developer Manual" |
| 2 | .SH NAME | |
| 05ec9ff7 DG |
3 | .B DEPRECATED |
| 4 | ||
| 2df048c2 DG |
5 | lttng_health_check \- Monitor health of the session daemon |
| 6 | .SH SYNOPSIS | |
| 7 | .nf | |
| 8 | .B #include <lttng/lttng.h> | |
| 9 | .sp | |
| 10 | .BI "int lttng_health_check(enum lttng_health_component c); | |
| 11 | .fi | |
| 12 | ||
| 13 | Link with -llttng-ctl. | |
| 14 | .SH DESCRIPTION | |
| 15 | The | |
| 16 | .BR lttng_health_check () | |
| 17 | is used to check the session daemon health for either a specific component | |
| 18 | .BR c | |
| 19 | or for all of them. Each component represent a subsystem of the session daemon. | |
| 20 | Those components are set with health counters that are atomically incremented | |
| 21 | once reached. An even value indicates progress in the execution of the | |
| 22 | component. An odd value means that the code has entered a blocking state which | |
| 23 | is not a poll(7) wait period. | |
| 24 | ||
| 25 | A bad health is defined by a fatal error code path reached or any IPC used in | |
| 26 | the session daemon that was blocked for more than 20 seconds (default timeout). | |
| 27 | The condition for this bad health to be detected is that one or many of the | |
| 28 | counters are odd. | |
| 29 | ||
| 30 | The health check mechanism of the session daemon can only be reached through | |
| 31 | the health socket which is a different one from the command and the application | |
| 32 | socket. An isolated thread serves this socket and only computes the health | |
| 33 | counters across the code when asked by the lttng control library (using this | |
| 34 | call). This subsystem is highly unlikely to fail due to its simplicity. | |
| 35 | ||
| 36 | The | |
| 37 | .BR c | |
| 38 | argument can be one of the following values: | |
| 39 | .TP | |
| 40 | .BR LTTNG_HEALTH_CMD | |
| 41 | Command subsystem which handles user commands coming from the liblttng-ctl or | |
| 42 | the | |
| 43 | .BR lttng(1) | |
| 44 | command line interface. | |
| 45 | .TP | |
| 46 | .BR LTTNG_HEALTH_APP_MANAGE | |
| 47 | The session daemon manages application socket in order to route client command | |
| 48 | and check if they get closed which indicates the application shutdown. | |
| 49 | .TP | |
| 50 | .BR LTTNG_HEALTH_APP_REG | |
| 51 | The application registration mechanism is an important and vital part of for | |
| 52 | user space tracing. Upon startup, applications instrumented with | |
| 53 | .BR lttng-ust(3) | |
| 54 | try to register to the session daemon through this subsystem. | |
| 55 | .TP | |
| 56 | .BR LTTNG_HEALTH_KERNEL | |
| 57 | Monitor the Kernel tracer streams and main channel of communication | |
| 58 | (/proc/lttng). If this component malfunction, the Kernel tracer is not usable | |
| 59 | anymore by lttng-tools. | |
| 60 | .TP | |
| 61 | .BR LTTNG_HEALTH_CONSUMER | |
| 62 | The session daemon can spawn up to | |
| 63 | .BR three | |
| 64 | consumer daemon for kernel, user space 32 and 64 bit. This subsystem monitors | |
| 65 | the consumer daemon(s). A bad health state means that the consumer(s) are not | |
| 66 | usable anymore hence likely making tracing not usable. | |
| 67 | .TP | |
| 68 | .BR LTTNG_HEALTH_ALL | |
| 69 | Check all components. If only one of them is in a bad state, a health check | |
| 70 | error is returned. | |
| 71 | ||
| 72 | .SH "RETURN VALUE" | |
| 73 | Return 0 if the health is OK, or 1 is it's in a bad state. A return code of \-1 | |
| 74 | indicates that the control library was not able to connect to the session | |
| 75 | daemon health socket. | |
| 76 | ||
| 77 | .SH "LIMITATIONS" | |
| 78 | ||
| 79 | For the LTTNG_HEALTH_CONSUMER, you can not know which consumer daemon has | |
| 80 | failed but only that either the consumer subsystem has failed or that a | |
| 81 | lttng-consumerd died. | |
| 82 | ||
| 83 | .SH "AUTHORS" | |
| 9b22d135 JG |
84 | lttng-health-check was originally written by David Goulet and is currently |
| 85 | maintained by Jérémie Galarneau <jeremie.galarneau@efficios.com>. |