From: Marc Fournier Date: Mon, 24 Nov 2014 08:44:45 +0000 (+0100) Subject: Merge branch 'collectd-5.4' X-Git-Tag: collectd-5.5.0~127 X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=e1812f8e5c5f91328398c66b602d07ee7c6ede51;p=thirdparty%2Fcollectd.git Merge branch 'collectd-5.4' Conflicts: src/Makefile.am src/cpu.c src/swap.c src/write_redis.c NB: dropped change to src/swap.c made in 1a146775d4, as suggested in the commit message. --- e1812f8e5c5f91328398c66b602d07ee7c6ede51 diff --cc src/collectd.conf.in index 5196979ea,1e0f78c17..a25cce55f --- a/src/collectd.conf.in +++ b/src/collectd.conf.in @@@ -1039,16 -1002,20 +1074,22 @@@ # DeleteGauges false # DeleteSets false # TimerPercentile 90.0 + # TimerPercentile 95.0 + # TimerPercentile 99.0 + # TimerLower false + # TimerUpper false + # TimerSum false + # TimerCount false # -# +# # ReportByDevice false # ReportBytes true +# ValuesAbsolute true +# ValuesPercentage false # -# +# # # Instance "slabinfo" # Separator " " diff --cc src/collectd.conf.pod index 20c612600,03a144432..a65d432bd --- a/src/collectd.conf.pod +++ b/src/collectd.conf.pod @@@ -7105,9 -6304,64 +7118,64 @@@ want to set B to =back + =head2 Plugin C + + The I submits values to I, a data structure server. + + Synopsis: + + + + Host "localhost" + Port "6379" + Timeout 1000 + + + + Values are submitted to I, using the metric name as the key, and + the timestamp as the score. Retrieving a date range can then be done using the + C I command. Additionnally, all the identifiers of these + I are kept in a I called C and can be + retrieved using the C I command. See + L and L for + details. + + The information shown in the synopsis above is the I + which is used by the plugin if no configuration is present. + + The plugin can send values to multiple instances of I by specifying + one B block for each instance. Within the B blocks, the following + options are available: + + =over 4 + + =item B I + + The B block identifies a new I node, that is a new I + instance running in an specified host and port. The name for node is a + canonical identifier which is used as I. It is limited to + 64Echaracters in length. + + =item B I + + The B option is the hostname or IP-address where the I instance is + running on. + + =item B I + + The B option is the TCP port on which the Redis instance accepts + connections. Either a service name of a port number may be given. Please note + that numerical port numbers must be given as a string, too. + + =item B I + + The B option sets the socket connection timeout, in milliseconds. + + =back + =head2 Plugin C -The I will send values to I, a powerfull stream +The I will send values to I, a powerful stream aggregation and monitoring system. The plugin sends I encoded data to I using UDP packets. diff --cc src/cpu.c index d1978f96b,1c4e5f6ba..2fb4918b9 --- a/src/cpu.c +++ b/src/cpu.c @@@ -172,57 -139,9 +166,57 @@@ static int numcpu static int pnumcpu; #endif /* HAVE_PERFSTAT */ +#define RATE_ADD(sum, val) do { \ + if (isnan (sum)) \ + (sum) = (val); \ + else if (!isnan (val)) \ + (sum) += (val); \ +} while (0) + +struct cpu_state_s +{ + value_to_rate_state_t conv; + gauge_t rate; + _Bool has_value; +}; +typedef struct cpu_state_s cpu_state_t; + +static cpu_state_t *cpu_states = NULL; +static size_t cpu_states_num = 0; /* #cpu_states allocated */ + +/* Highest CPU number in the current iteration. Used by the dispatch logic to + * determine how many CPUs there were. Reset to 0 by cpu_reset(). */ +static size_t global_cpu_num = 0; + +static _Bool report_by_cpu = 1; +static _Bool report_by_state = 1; +static _Bool report_percent = 0; + +static const char *config_keys[] = +{ + "ReportByCpu", + "ReportByState", + "ValuesPercentage" +}; +static int config_keys_num = STATIC_ARRAY_SIZE (config_keys); + +static int cpu_config (char const *key, char const *value) /* {{{ */ +{ + if (strcasecmp (key, "ReportByCpu") == 0) + report_by_cpu = IS_TRUE (value) ? 1 : 0; + else if (strcasecmp (key, "ValuesPercentage") == 0) + report_percent = IS_TRUE (value) ? 1 : 0; + else if (strcasecmp (key, "ReportByState") == 0) + report_by_state = IS_TRUE (value) ? 1 : 0; + else + return (-1); + + return (0); +} /* }}} int cpu_config */ + static int init (void) { - #if PROCESSOR_CPU_LOAD_INFO || PROCESSOR_TEMPERATURE + #if PROCESSOR_CPU_LOAD_INFO kern_return_t status; port_host = mach_host_self (); @@@ -339,257 -254,15 +333,251 @@@ static void submit_value (int cpu_num, plugin_dispatch_values (&vl); } +static void submit_percent(int cpu_num, int cpu_state, gauge_t percent) +{ + value_t value; + + /* This function is called for all known CPU states, but each read + * method will only report a subset. The remaining states are left as + * NAN and we ignore them here. */ + if (isnan (percent)) + return; + + value.gauge = percent; + submit_value (cpu_num, cpu_state, "percent", value); +} + +static void submit_derive(int cpu_num, int cpu_state, derive_t derive) +{ + value_t value; + + value.derive = derive; + submit_value (cpu_num, cpu_state, "cpu", value); +} + +/* Takes the zero-index number of a CPU and makes sure that the module-global + * cpu_states buffer is large enough. Returne ENOMEM on erorr. */ +static int cpu_states_alloc (size_t cpu_num) /* {{{ */ +{ + cpu_state_t *tmp; + size_t sz; + + sz = (((size_t) cpu_num) + 1) * CPU_STATE_MAX; + assert (sz > 0); + + /* We already have enough space. */ + if (cpu_states_num >= sz) + return 0; + + tmp = realloc (cpu_states, sz * sizeof (*cpu_states)); + if (tmp == NULL) + { + ERROR ("cpu plugin: realloc failed."); + return (ENOMEM); + } + cpu_states = tmp; + tmp = cpu_states + cpu_states_num; + + memset (tmp, 0, (sz - cpu_states_num) * sizeof (*cpu_states)); + cpu_states_num = sz; + return 0; +} /* }}} cpu_states_alloc */ + +static cpu_state_t *get_cpu_state (size_t cpu_num, size_t state) /* {{{ */ +{ + size_t index = ((cpu_num * CPU_STATE_MAX) + state); + + if (index >= cpu_states_num) + return (NULL); + + return (&cpu_states[index]); +} /* }}} cpu_state_t *get_cpu_state */ + +/* Populates the per-CPU CPU_STATE_ACTIVE rate and the global rate_by_state + * array. */ +static void aggregate (gauge_t *sum_by_state) /* {{{ */ +{ + size_t cpu_num; + size_t state; + + for (state = 0; state < CPU_STATE_MAX; state++) + sum_by_state[state] = NAN; + + for (cpu_num = 0; cpu_num < global_cpu_num; cpu_num++) + { + cpu_state_t *this_cpu_states = get_cpu_state (cpu_num, 0); + + this_cpu_states[CPU_STATE_ACTIVE].rate = NAN; + + for (state = 0; state < CPU_STATE_ACTIVE; state++) + { + if (!this_cpu_states[state].has_value) + continue; + + RATE_ADD (sum_by_state[state], this_cpu_states[state].rate); + if (state != CPU_STATE_IDLE) + RATE_ADD (this_cpu_states[CPU_STATE_ACTIVE].rate, this_cpu_states[state].rate); + } + + RATE_ADD (sum_by_state[CPU_STATE_ACTIVE], this_cpu_states[CPU_STATE_ACTIVE].rate); + } +} /* }}} void aggregate */ + +/* Commits (dispatches) the values for one CPU or the global aggregation. + * cpu_num is the index of the CPU to be committed or -1 in case of the global + * aggregation. rates is a pointer to CPU_STATE_MAX gauge_t values holding the + * current rate; each rate may be NAN. Calculates the percentage of each state + * and dispatches the metric. */ +static void cpu_commit_one (int cpu_num, /* {{{ */ + gauge_t rates[static CPU_STATE_MAX]) +{ + size_t state; + gauge_t sum; + + sum = rates[CPU_STATE_ACTIVE]; + RATE_ADD (sum, rates[CPU_STATE_IDLE]); + + if (!report_by_state) + { + gauge_t percent = 100.0 * rates[CPU_STATE_ACTIVE] / sum; + submit_percent (cpu_num, CPU_STATE_ACTIVE, percent); + return; + } + + for (state = 0; state < CPU_STATE_ACTIVE; state++) + { + gauge_t percent = 100.0 * rates[state] / sum; + submit_percent (cpu_num, state, percent); + } +} /* }}} void cpu_commit_one */ + +/* Resets the internal aggregation. This is called by the read callback after + * each iteration / after each call to cpu_commit(). */ +static void cpu_reset (void) /* {{{ */ +{ + size_t i; + + for (i = 0; i < cpu_states_num; i++) + cpu_states[i].has_value = 0; + + global_cpu_num = 0; +} /* }}} void cpu_reset */ + +/* Legacy behavior: Dispatches the raw derive values without any aggregation. */ +static void cpu_commit_without_aggregation (void) /* {{{ */ +{ + int state; + + for (state = 0; state < CPU_STATE_ACTIVE; state++) + { + size_t cpu_num; + if (report_by_cpu) { + for (cpu_num = 0; cpu_num < global_cpu_num; cpu_num++) + { + cpu_state_t *s = get_cpu_state (cpu_num, state); + + if (!s->has_value) + continue; + + submit_derive ((int) cpu_num, (int) state, s->conv.last_value.derive); + } + } else { + derive_t derive_total = 0; + for (cpu_num = 0; cpu_num < global_cpu_num; cpu_num++) + { + cpu_state_t *s = get_cpu_state (cpu_num, state); + + if (!s->has_value) + continue; + + derive_total += s->conv.last_value.derive; + + } + submit_derive (-1, (int) state, derive_total); + } + } +} /* }}} void cpu_commit_without_aggregation */ + +/* Aggregates the internal state and dispatches the metrics. */ +static void cpu_commit (void) /* {{{ */ +{ + gauge_t global_rates[CPU_STATE_MAX] = { + NAN, NAN, NAN, NAN, NAN, NAN, NAN, NAN, NAN, NAN + }; + size_t cpu_num; + + if (report_by_state && !report_percent) + { + cpu_commit_without_aggregation (); + return; + } + + aggregate (global_rates); + + if (!report_by_cpu) + { + cpu_commit_one (-1, global_rates); + return; + } + + for (cpu_num = 0; cpu_num < global_cpu_num; cpu_num++) + { + cpu_state_t *this_cpu_states = get_cpu_state (cpu_num, 0); + gauge_t local_rates[CPU_STATE_MAX] = { + NAN, NAN, NAN, NAN, NAN, NAN, NAN, NAN, NAN, NAN + }; + size_t state; + + for (state = 0; state < CPU_STATE_ACTIVE; state++) + if (this_cpu_states[state].has_value) + local_rates[state] = this_cpu_states[state].rate; + + cpu_commit_one ((int) cpu_num, local_rates); + } +} /* }}} void cpu_commit */ + +/* Adds a derive value to the internal state. This should be used by each read + * function for each state. At the end of the iteration, the read function + * should call cpu_commit(). */ +static int cpu_stage (size_t cpu_num, size_t state, derive_t value, cdtime_t now) /* {{{ */ +{ + int status; + cpu_state_t *s; + value_t v; + + if (state >= CPU_STATE_ACTIVE) + return (EINVAL); + + status = cpu_states_alloc (cpu_num); + if (status != 0) + return (status); + + if (global_cpu_num <= cpu_num) + global_cpu_num = cpu_num + 1; + + s = get_cpu_state (cpu_num, state); + + v.gauge = NAN; + status = value_to_rate (&v, value, &s->conv, DS_TYPE_DERIVE, now); + if (status != 0) + return (status); + + s->rate = v.gauge; + s->has_value = 1; + return (0); +} /* }}} int cpu_stage */ + static int cpu_read (void) { -#if PROCESSOR_CPU_LOAD_INFO + cdtime_t now = cdtime (); + - #if PROCESSOR_CPU_LOAD_INFO || PROCESSOR_TEMPERATURE /* {{{ */ ++#if PROCESSOR_CPU_LOAD_INFO /* {{{ */ int cpu; kern_return_t status; - + - #if PROCESSOR_CPU_LOAD_INFO processor_cpu_load_info_data_t cpu_info; mach_msg_type_number_t cpu_info_len; - #endif - #if PROCESSOR_TEMPERATURE - processor_info_data_t cpu_temp; - mach_msg_type_number_t cpu_temp_len; - #endif host_t cpu_host; @@@ -613,59 -286,17 +601,15 @@@ continue; } - submit (cpu, "user", (derive_t) cpu_info.cpu_ticks[CPU_STATE_USER]); - submit (cpu, "nice", (derive_t) cpu_info.cpu_ticks[CPU_STATE_NICE]); - submit (cpu, "system", (derive_t) cpu_info.cpu_ticks[CPU_STATE_SYSTEM]); - submit (cpu, "idle", (derive_t) cpu_info.cpu_ticks[CPU_STATE_IDLE]); + cpu_stage (cpu, CPU_STATE_USER, (derive_t) cpu_info.cpu_ticks[CPU_STATE_USER], now); + cpu_stage (cpu, CPU_STATE_NICE, (derive_t) cpu_info.cpu_ticks[CPU_STATE_NICE], now); + cpu_stage (cpu, CPU_STATE_SYSTEM, (derive_t) cpu_info.cpu_ticks[CPU_STATE_SYSTEM], now); + cpu_stage (cpu, CPU_STATE_IDLE, (derive_t) cpu_info.cpu_ticks[CPU_STATE_IDLE], now); - #endif /* PROCESSOR_CPU_LOAD_INFO */ - - #if PROCESSOR_TEMPERATURE - /* - * Not all Apple computers do have this ability. To minimize - * the messages sent to the syslog we do an exponential - * stepback if `processor_info' fails. We still try ~once a day - * though.. - */ - if (cpu_temp_retry_counter > 0) - { - cpu_temp_retry_counter--; - continue; - } - - cpu_temp_len = PROCESSOR_INFO_MAX; - - status = processor_info (cpu_list[cpu], - PROCESSOR_TEMPERATURE, - &cpu_host, - cpu_temp, &cpu_temp_len); - if (status != KERN_SUCCESS) - { - ERROR ("cpu plugin: processor_info failed: %s", - mach_error_string (status)); - - cpu_temp_retry_counter = cpu_temp_retry_step; - cpu_temp_retry_step *= 2; - if (cpu_temp_retry_step > cpu_temp_retry_max) - cpu_temp_retry_step = cpu_temp_retry_max; - - continue; - } - - if (cpu_temp_len != 1) - { - DEBUG ("processor_info (PROCESSOR_TEMPERATURE) returned %i elements..?", - (int) cpu_temp_len); - continue; - } - - cpu_temp_retry_counter = 0; - cpu_temp_retry_step = 1; - #endif /* PROCESSOR_TEMPERATURE */ } -/* #endif PROCESSOR_CPU_LOAD_INFO */ +/* }}} #endif PROCESSOR_CPU_LOAD_INFO */ -#elif defined(KERNEL_LINUX) +#elif defined(KERNEL_LINUX) /* {{{ */ int cpu; - derive_t user, nice, syst, idle; - derive_t wait, intr, sitr; /* sitr == soft interrupt */ FILE *fh; char buf[1024];