X-Git-Url: https://git.octo.it/?a=blobdiff_plain;f=src%2Flpar.c;h=412a721ab3707f3d448f13692599e2fc705855f6;hb=678d8eafdf17da0ef94053f5c017c12152eb69b0;hp=7998d7963202a4dc7e802d72dbe9510e108a9765;hpb=f3feef5de90c0c926ee76c96972c4ce4b8b5c838;p=collectd.git diff --git a/src/lpar.c b/src/lpar.c index 7998d796..412a721a 100644 --- a/src/lpar.c +++ b/src/lpar.c @@ -16,242 +16,237 @@ * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA * * Authors: - * Aurelien Reynaud + * Aurélien Reynaud **/ #include "collectd.h" + #include "common.h" #include "plugin.h" -#include #include -#include +#include #include +/* XINTFRAC was defined in libperfstat.h somewhere between AIX 5.3 and 6.1 */ #ifndef XINTFRAC -# define XINTFRAC ((double)(_system_configuration.Xint) / \ - (double)(_system_configuration.Xfrac)) +#include +#define XINTFRAC \ + ((double)(_system_configuration.Xint) / (double)(_system_configuration.Xfrac)) #endif -static const char *config_keys[] = -{ - "CpuPoolStats" -}; -static int config_keys_num = STATIC_ARRAY_SIZE (config_keys); -static int pool_stats = 0; - -/* As an LPAR can be moved transparently across physical systems - * through Live Partition Mobility (LPM), and the resources we are - * monitoring are tied to the underlying hardware, we need to keep - * track on which physical server we are currently on. This is done - * through the plugin instance which holds the chassis' serial. - */ -static u_longlong_t last_time_base; -static u_longlong_t last_pcpu_user, - last_pcpu_sys, - last_pcpu_idle, - last_pcpu_wait; -static u_longlong_t last_pool_idle_time = 0; -static u_longlong_t last_idle_donated_purr = 0, - last_busy_donated_purr = 0, - last_busy_stolen_purr = 0, - last_idle_stolen_purr = 0; -static int donate_flag = 0; - - -/* Save the current values for the next iteration */ -static void save_last_values (perfstat_partition_total_t *lparstats) -{ - last_time_base = lparstats->timebase_last; - - last_pcpu_user = lparstats->puser; - last_pcpu_sys = lparstats->psys; - last_pcpu_idle = lparstats->pidle; - last_pcpu_wait = lparstats->pwait; - - if (donate_flag) - { - last_idle_donated_purr = lparstats->idle_donated_purr; - last_busy_donated_purr = lparstats->busy_donated_purr; - last_busy_stolen_purr = lparstats->busy_stolen_purr; - last_idle_stolen_purr = lparstats->idle_stolen_purr; - } - - last_pool_idle_time = lparstats->pool_idle_time; -} - -static int lpar_config (const char *key, const char *value) -{ - if (strcasecmp ("CpuPoolStats", key) == 0) - { - if (IS_TRUE (value)) - pool_stats = 1; - else - pool_stats = 0; - } - else - { - return (-1); - } - - return (0); -} /* int lpar_config */ +#define CLOCKTICKS_TO_TICKS(cticks) ((cticks) / XINTFRAC) -static int lpar_init (void) -{ - perfstat_partition_total_t lparstats; +static const char *config_keys[] = {"CpuPoolStats", "ReportBySerial"}; +static int config_keys_num = STATIC_ARRAY_SIZE(config_keys); - /* retrieve the initial metrics */ - if (!perfstat_partition_total (NULL, &lparstats, - sizeof (perfstat_partition_total_t), 1)) - { - ERROR ("lpar plugin: perfstat_partition_total failed."); - return (-1); - } +static _Bool pool_stats = 0; +static _Bool report_by_serial = 0; +#if PERFSTAT_SUPPORTS_DONATION +static _Bool donate_flag = 0; +#endif +static char serial[SYS_NMLN]; + +static perfstat_partition_total_t lparstats_old; + +static int lpar_config(const char *key, const char *value) { + if (strcasecmp("CpuPoolStats", key) == 0) { + if (IS_TRUE(value)) + pool_stats = 1; + else + pool_stats = 0; + } else if (strcasecmp("ReportBySerial", key) == 0) { + if (IS_TRUE(value)) + report_by_serial = 1; + else + report_by_serial = 0; + } else { + return (-1); + } + + return (0); +} /* int lpar_config */ - if (!lparstats.type.b.shared_enabled && lparstats.type.b.donate_enabled) - { - donate_flag = 1; - } +static int lpar_init(void) { + int status; + + /* Retrieve the initial metrics. Returns the number of structures filled. */ + status = perfstat_partition_total(/* name = */ NULL, /* (must be NULL) */ + &lparstats_old, + sizeof(perfstat_partition_total_t), + /* number = */ 1 /* (must be 1) */); + if (status != 1) { + char errbuf[1024]; + ERROR("lpar plugin: perfstat_partition_total failed: %s (%i)", + sstrerror(errno, errbuf, sizeof(errbuf)), status); + return (-1); + } + +#if PERFSTAT_SUPPORTS_DONATION + if (!lparstats_old.type.b.shared_enabled && + lparstats_old.type.b.donate_enabled) { + donate_flag = 1; + } +#endif - /* save the initial data */ - save_last_values (&lparstats); + if (pool_stats && !lparstats_old.type.b.pool_util_authority) { + WARNING("lpar plugin: This partition does not have pool authority. " + "Disabling CPU pool statistics collection."); + pool_stats = 0; + } - return (0); + return (0); } /* int lpar_init */ -static void lpar_submit (const char *plugin_inst, const char *type_instance, double value) -{ - value_t values[1]; - value_list_t vl = VALUE_LIST_INIT; - - values[0].gauge = (gauge_t)value; - - vl.values = values; - vl.values_len = 1; - sstrncpy (vl.host, hostname_g, sizeof (vl.host)); - sstrncpy (vl.plugin, "lpar", sizeof (vl.plugin)); - sstrncpy (vl.plugin_instance, plugin_inst, sizeof (vl.plugin)); - sstrncpy (vl.type, "lpar_pcpu", sizeof (vl.type)); - sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance)); - - plugin_dispatch_values (&vl); -} - -static int submit_counter (const char *plugin_instance, /* {{{ */ - const char *type, const char *type_instance, counter_t value) -{ - value_t values[1]; - value_list_t vl = VALUE_LIST_INIT; - - values[0].counter = value; - - vl.values = values; - vl.values_len = 1; - sstrncpy (vl.host, hostname_g, sizeof (vl.host)); - sstrncpy (vl.plugin, "lpar", sizeof (vl.plugin)); - sstrncpy (vl.plugin_instance, plugin_inst, sizeof (vl.plugin)); - sstrncpy (vl.type, type, sizeof (vl.type)); - sstrncpy (vl.type_instance, type_instance, sizeof (vl.type_instance)); - - return (plugin_dispatch_values (&vl)); -} /* }}} int submit_counter */ - -static int lpar_read (void) -{ - u_longlong_t delta_time_base; - perfstat_partition_total_t lparstats; - struct utsname name; - char plugin_inst[DATA_MAX_NAME_LEN]; - _Bool have_donate = 0; - - /* retrieve the current physical server's id and build the plugin - instance's name */ - if (uname (&name) != 0) - { - ERROR ("lpar plugin: uname failed."); - return (-1); - } - sstrncpy (plugin_inst, name.machine, sizeof (plugin_inst)); - - /* retrieve the current metrics */ - if (!perfstat_partition_total (/* name = */ NULL, /* "must be set to NULL" */ - &lparstats, sizeof (lparstats), - /* desired_number = */ 1 /* "must be set to 1" */)) - { - ERROR ("lpar plugin: perfstat_partition_total failed."); - return (-1); - } - - if (!lparstats.type.b.shared_enabled - && lparstats.type.b.donate_enabled) - have_donate = 1; - - delta_time_base = lparstats.timebase_last - last_time_base; - if (delta_time_base == 0) - { - /* The system stats have not been updated since last time */ - return (0); - } - - submit_counter (plugin_inst, "cpu", "user", (counter_t) lparstats.puser); - submit_counter (plugin_inst, "cpu", "system", (counter_t) lparstats.psys); - submit_counter (plugin_inst, "cpu", "idle", (counter_t) lparstats.pidle); - submit_counter (plugin_inst, "cpu", "wait", (counter_t) lparstats.pwait); - - /* FIXME: Use an appropriate GAUGE type here. */ - lpar_submit (plugin_inst, "ent", (double)lparstats.entitled_proc_capacity / 100.0); - lpar_submit (plugin_inst, "max", (double)lparstats.max_proc_capacity / 100.0); - lpar_submit (plugin_inst, "min", (double)lparstats.min_proc_capacity / 100.0); - - if (have_donate) - { - dlt_idle_donated = lparstats.idle_donated_purr - last_idle_donated_purr; - dlt_busy_donated = lparstats.busy_donated_purr - last_busy_donated_purr; - dlt_idle_stolen = lparstats.idle_stolen_purr - last_idle_stolen_purr; - dlt_busy_stolen = lparstats.busy_stolen_purr - last_busy_stolen_purr; - - submit_counter (plugin_inst, "cpu", "donated-idle", (counter_t) lparstats.idle_donated_purr); - submit_counter (plugin_inst, "cpu", "donated-busy", (counter_t) lparstats.busy_donated_purr); - submit_counter (plugin_inst, "cpu", "stolen-idle", (counter_t) lparstats.idle_stolen_purr); - submit_counter (plugin_inst, "cpu", "stolen-busy", (counter_t) lparstats.busy_stolen_purr); - } - - if (pool_stats) - { - if (!lparstats.type.b.pool_util_authority) - { - WARNING ("lpar plugin: Pool utilization data is not available."); - } - else - { - u_longlong_t dlt_pit; - double total, idle; - char type[DATA_MAX_NAME_LEN]; - - /* FIXME: The pool id should probably be used as plugin instance. */ - dlt_pit = lparstats.pool_idle_time - last_pool_idle_time; - total = (double)lparstats.phys_cpus_pool; - idle = (double)dlt_pit / XINTFRAC / (double)delta_time_base; - ssnprintf (type, sizeof(type), "pool-%X-total", lparstats.pool_id); - lpar_submit (plugin_inst, type, total); - ssnprintf (type, sizeof(type), "pool-%X-used", lparstats.pool_id); - lpar_submit (plugin_inst, type, total - idle); - } - } - - save_last_values (&lparstats); - - return (0); +static void lpar_submit(const char *type_instance, double value) { + value_t values[1]; + value_list_t vl = VALUE_LIST_INIT; + + values[0].gauge = (gauge_t)value; + + vl.values = values; + vl.values_len = 1; + if (report_by_serial) { + sstrncpy(vl.host, serial, sizeof(vl.host)); + sstrncpy(vl.plugin_instance, hostname_g, sizeof(vl.plugin)); + } else { + sstrncpy(vl.host, hostname_g, sizeof(vl.host)); + } + sstrncpy(vl.plugin, "lpar", sizeof(vl.plugin)); + sstrncpy(vl.type, "vcpu", sizeof(vl.type)); + sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance)); + + plugin_dispatch_values(&vl); +} /* void lpar_submit */ + +static int lpar_read(void) { + perfstat_partition_total_t lparstats; + int status; + struct utsname name; + u_longlong_t ticks; + u_longlong_t user_ticks, syst_ticks, wait_ticks, idle_ticks; + u_longlong_t consumed_ticks; + double entitled_proc_capacity; + + /* An LPAR has the same serial number as the physical system it is currently + running on. It is a convenient way of tracking LPARs as they are moved + from chassis to chassis through Live Partition Mobility (LPM). */ + if (uname(&name) != 0) { + ERROR("lpar plugin: uname failed."); + return (-1); + } + sstrncpy(serial, name.machine, sizeof(serial)); + + /* Retrieve the current metrics. Returns the number of structures filled. */ + status = + perfstat_partition_total(/* name = */ NULL, /* (must be NULL) */ + &lparstats, sizeof(perfstat_partition_total_t), + /* number = */ 1 /* (must be 1) */); + if (status != 1) { + char errbuf[1024]; + ERROR("lpar plugin: perfstat_partition_total failed: %s (%i)", + sstrerror(errno, errbuf, sizeof(errbuf)), status); + return (-1); + } + + /* Number of ticks since we last run. */ + ticks = lparstats.timebase_last - lparstats_old.timebase_last; + if (ticks == 0) { + /* The stats have not been updated. Return now to avoid + * dividing by zero */ + return (0); + } + + /* + * On a shared partition, we're "entitled" to a certain amount of + * processing power, for example 250/100 of a physical CPU. Processing + * capacity not used by the partition may be assigned to a different + * partition by the hypervisor, so "idle" is hopefully a very small + * number. + * + * A dedicated partition may donate its CPUs to another partition and + * may steal ticks from somewhere else (another partition or maybe the + * shared pool, I don't know --octo). + */ + + /* entitled_proc_capacity is in 1/100th of a CPU */ + entitled_proc_capacity = 0.01 * ((double)lparstats.entitled_proc_capacity); + lpar_submit("entitled", entitled_proc_capacity); + + /* The number of ticks actually spent in the various states */ + user_ticks = lparstats.puser - lparstats_old.puser; + syst_ticks = lparstats.psys - lparstats_old.psys; + wait_ticks = lparstats.pwait - lparstats_old.pwait; + idle_ticks = lparstats.pidle - lparstats_old.pidle; + consumed_ticks = user_ticks + syst_ticks + wait_ticks + idle_ticks; + + lpar_submit("user", (double)user_ticks / (double)ticks); + lpar_submit("system", (double)syst_ticks / (double)ticks); + lpar_submit("wait", (double)wait_ticks / (double)ticks); + lpar_submit("idle", (double)idle_ticks / (double)ticks); + +#if PERFSTAT_SUPPORTS_DONATION + if (donate_flag) { + /* donated => ticks given to another partition + * stolen => ticks received from another partition */ + u_longlong_t idle_donated_ticks, busy_donated_ticks; + u_longlong_t idle_stolen_ticks, busy_stolen_ticks; + + /* FYI: PURR == Processor Utilization of Resources Register + * SPURR == Scaled PURR */ + idle_donated_ticks = + lparstats.idle_donated_purr - lparstats_old.idle_donated_purr; + busy_donated_ticks = + lparstats.busy_donated_purr - lparstats_old.busy_donated_purr; + idle_stolen_ticks = + lparstats.idle_stolen_purr - lparstats_old.idle_stolen_purr; + busy_stolen_ticks = + lparstats.busy_stolen_purr - lparstats_old.busy_stolen_purr; + + lpar_submit("idle_donated", (double)idle_donated_ticks / (double)ticks); + lpar_submit("busy_donated", (double)busy_donated_ticks / (double)ticks); + lpar_submit("idle_stolen", (double)idle_stolen_ticks / (double)ticks); + lpar_submit("busy_stolen", (double)busy_stolen_ticks / (double)ticks); + + /* Donated ticks will be accounted for as stolen ticks in other LPARs */ + consumed_ticks += idle_stolen_ticks + busy_stolen_ticks; + } +#endif + + lpar_submit("consumed", (double)consumed_ticks / (double)ticks); + + if (pool_stats) { + char typinst[DATA_MAX_NAME_LEN]; + u_longlong_t pool_idle_cticks; + double pool_idle_cpus; + double pool_busy_cpus; + + /* We're calculating "busy" from "idle" and the total number of + * CPUs, because the "busy" member didn't exist in early versions + * of libperfstat. It was added somewhere between AIX 5.3 ML5 and ML9. */ + pool_idle_cticks = lparstats.pool_idle_time - lparstats_old.pool_idle_time; + pool_idle_cpus = + CLOCKTICKS_TO_TICKS((double)pool_idle_cticks) / (double)ticks; + pool_busy_cpus = ((double)lparstats.phys_cpus_pool) - pool_idle_cpus; + if (pool_busy_cpus < 0.0) + pool_busy_cpus = 0.0; + + ssnprintf(typinst, sizeof(typinst), "pool-%X-busy", lparstats.pool_id); + lpar_submit(typinst, pool_busy_cpus); + + ssnprintf(typinst, sizeof(typinst), "pool-%X-idle", lparstats.pool_id); + lpar_submit(typinst, pool_idle_cpus); + } + + memcpy(&lparstats_old, &lparstats, sizeof(lparstats_old)); + + return (0); } /* int lpar_read */ -void module_register (void) -{ - plugin_register_config ("lpar", lpar_config, - config_keys, config_keys_num); - plugin_register_init ("lpar", lpar_init); - plugin_register_read ("lpar", lpar_read); +void module_register(void) { + plugin_register_config("lpar", lpar_config, config_keys, config_keys_num); + plugin_register_init("lpar", lpar_init); + plugin_register_read("lpar", lpar_read); } /* void module_register */ -/* vim: set sw=8 sts=8 ts=8 noet : */ - +/* vim: set sw=8 noet : */