src/common/health/health.c

   1 /*
   2  * Copyright (C) 2012 - David Goulet <dgoulet@efficios.com>
   3  * Copyright (C) 2013 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License, version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along with
  15  * this program; if not, write to the Free Software Foundation, Inc., 51
  16  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _LGPL_SOURCE
  20 #include <assert.h>
  21 #include <inttypes.h>
  22 #include <stdio.h>
  23 #include <stdlib.h>
  24 #include <time.h>
  25
  26 #include <common/defaults.h>
  27 #include <common/error.h>
  28 #include <common/macros.h>
  29 #include <common/sessiond-comm/inet.h>
  30
  31 #include <lttng/health-internal.h>
  32
  33 /*
  34  * An application-specific error state for unregistered thread keeps
  35  * track of thread errors. A thread reporting a health error, normally
  36  * unregisters and quits. This makes the TLS health state not available
  37  * to the health_check_state() call so on unregister we update this
  38  * global error array so we can keep track of which thread was on error
  39  * if the TLS health state has been removed.
  40  */
  41 struct health_app {
  42         /* List of health state, for each application thread */
  43         struct cds_list_head list;
  44         /*
  45          * This lock ensures that TLS memory used for the node and its
  46          * container structure don't get reclaimed after the TLS owner
  47          * thread exits until we have finished using it.
  48          */
  49         pthread_mutex_t lock;
  50         int nr_types;
  51         struct timespec time_delta;
  52         /* Health flags containing thread type error state */
  53         enum health_flags *flags;
  54 };
  55
  56 /* Define TLS health state. */
  57 DEFINE_URCU_TLS(struct health_state, health_state);
  58
  59 /*
  60  * Initialize health check subsytem.
  61  */
  62 static
  63 void health_init(struct health_app *ha)
  64 {
  65         /*
  66          * Get the maximum value between the default delta value and the TCP
  67          * timeout with a safety net of the default health check delta.
  68          */
  69         ha->time_delta.tv_sec = max_t(unsigned long,
  70                         lttcomm_inet_tcp_timeout + DEFAULT_HEALTH_CHECK_DELTA_S,
  71                         ha->time_delta.tv_sec);
  72         DBG("Health check time delta in seconds set to %lu",
  73                 ha->time_delta.tv_sec);
  74 }
  75
  76 struct health_app *health_app_create(int nr_types)
  77 {
  78         struct health_app *ha;
  79
  80         ha = zmalloc(sizeof(*ha));
  81         if (!ha) {
  82                 return NULL;
  83         }
  84         ha->flags = zmalloc(sizeof(*ha->flags) * nr_types);
  85         if (!ha->flags) {
  86                 goto error_flags;
  87         }
  88         CDS_INIT_LIST_HEAD(&ha->list);
  89         pthread_mutex_init(&ha->lock, NULL);
  90         ha->nr_types = nr_types;
  91         ha->time_delta.tv_sec = DEFAULT_HEALTH_CHECK_DELTA_S;
  92         ha->time_delta.tv_nsec = DEFAULT_HEALTH_CHECK_DELTA_NS;
  93         health_init(ha);
  94         return ha;
  95
  96 error_flags:
  97         free(ha);
  98         return NULL;
  99 }
 100
 101 void health_app_destroy(struct health_app *ha)
 102 {
 103         free(ha->flags);
 104         free(ha);
 105 }
 106
 107 /*
 108  * Lock health state global list mutex.
 109  */
 110 static void state_lock(struct health_app *ha)
 111 {
 112         pthread_mutex_lock(&ha->lock);
 113 }
 114
 115 /*
 116  * Unlock health state global list mutex.
 117  */
 118 static void state_unlock(struct health_app *ha)
 119 {
 120         pthread_mutex_unlock(&ha->lock);
 121 }
 122
 123 /*
 124  * Set time difference in res from time_a and time_b.
 125  */
 126 static void time_diff(const struct timespec *time_a,
 127                 const struct timespec *time_b, struct timespec *res)
 128 {
 129         if (time_a->tv_nsec - time_b->tv_nsec < 0) {
 130                 res->tv_sec = time_a->tv_sec - time_b->tv_sec - 1;
 131                 res->tv_nsec = 1000000000L + time_a->tv_sec - time_b->tv_sec;
 132         } else {
 133                 res->tv_sec = time_a->tv_sec - time_b->tv_sec;
 134                 res->tv_nsec = time_a->tv_nsec - time_b->tv_nsec;
 135         }
 136 }
 137
 138 /*
 139  * Return true if time_a - time_b > diff, else false.
 140  */
 141 static int time_diff_gt(const struct timespec *time_a,
 142                 const struct timespec *time_b, const struct timespec *diff)
 143 {
 144         struct timespec res;
 145
 146         time_diff(time_a, time_b, &res);
 147         time_diff(&res, diff, &res);
 148
 149         if (res.tv_sec > 0) {
 150                 return 1;
 151         } else if (res.tv_sec == 0 && res.tv_nsec > 0) {
 152                 return 1;
 153         }
 154
 155         return 0;
 156 }
 157
 158 /*
 159  * Validate health state. Checks for the error flag or health conditions.
 160  *
 161  * Return 0 if health is bad or else 1.
 162  */
 163 static int validate_state(struct health_app *ha, struct health_state *state)
 164 {
 165         int retval = 1, ret;
 166         unsigned long current, last;
 167         struct timespec current_time;
 168
 169         assert(state);
 170
 171         last = state->last;
 172         current = uatomic_read(&state->current);
 173
 174         ret = clock_gettime(CLOCK_MONOTONIC, &current_time);
 175         if (ret < 0) {
 176                 PERROR("Error reading time\n");
 177                 /* error */
 178                 retval = 0;
 179                 goto end;
 180         }
 181
 182         /*
 183          * Thread is in bad health if flag HEALTH_ERROR is set. It is also in bad
 184          * health if, after the delta delay has passed, its the progress counter
 185          * has not moved and it has NOT been waiting for a poll() call.
 186          */
 187         if (uatomic_read(&state->flags) & HEALTH_ERROR) {
 188                 retval = 0;
 189                 goto end;
 190         }
 191
 192         /*
 193          * Initial condition need to update the last counter and sample time, but
 194          * should not check health in this initial case, because we don't know how
 195          * much time has passed.
 196          */
 197         if (state->last_time.tv_sec == 0 && state->last_time.tv_nsec == 0) {
 198                 /* update last counter and last sample time */
 199                 state->last = current;
 200                 memcpy(&state->last_time, &current_time, sizeof(current_time));
 201         } else {
 202                 if (time_diff_gt(&current_time, &state->last_time,
 203                                 &ha->time_delta)) {
 204                         if (current == last && !HEALTH_IS_IN_POLL(current)) {
 205                                 /* error */
 206                                 retval = 0;
 207                         }
 208                         /* update last counter and last sample time */
 209                         state->last = current;
 210                         memcpy(&state->last_time, &current_time, sizeof(current_time));
 211
 212                         /* On error, stop right now and notify caller. */
 213                         if (retval == 0) {
 214                                 goto end;
 215                         }
 216                 }
 217         }
 218
 219 end:
 220         DBG("Health state current %lu, last %lu, ret %d",
 221                         current, last, ret);
 222         return retval;
 223 }
 224
 225 /*
 226  * Check health of a specific health type. Note that if a thread has not yet
 227  * initialize its health subsystem or has quit, it's considered in a good
 228  * state.
 229  *
 230  * Return 0 if health is bad or else 1.
 231  */
 232 int health_check_state(struct health_app *ha, int type)
 233 {
 234         int retval = 1;
 235         struct health_state *state;
 236
 237         assert(type < ha->nr_types);
 238
 239         state_lock(ha);
 240
 241         cds_list_for_each_entry(state, &ha->list, node) {
 242                 int ret;
 243
 244                 if (state->type != type) {
 245                         continue;
 246                 }
 247
 248                 ret = validate_state(ha, state);
 249                 if (!ret) {
 250                         retval = 0;
 251                         goto end;
 252                 }
 253         }
 254
 255         /* Check the global state since some state might not be visible anymore. */
 256         if (ha->flags[type] & HEALTH_ERROR) {
 257                 retval = 0;
 258         }
 259
 260 end:
 261         state_unlock(ha);
 262
 263         DBG("Health check for type %d is %s", (int) type,
 264                         (retval == 0) ? "BAD" : "GOOD");
 265         return retval;
 266 }
 267
 268 /*
 269  * Init health state.
 270  */
 271 void health_register(struct health_app *ha, int type)
 272 {
 273         assert(type < ha->nr_types);
 274
 275         /* Init TLS state. */
 276         uatomic_set(&URCU_TLS(health_state).last, 0);
 277         uatomic_set(&URCU_TLS(health_state).last_time.tv_sec, 0);
 278         uatomic_set(&URCU_TLS(health_state).last_time.tv_nsec, 0);
 279         uatomic_set(&URCU_TLS(health_state).current, 0);
 280         uatomic_set(&URCU_TLS(health_state).flags, 0);
 281         uatomic_set(&URCU_TLS(health_state).type, type);
 282
 283         /* Add it to the global TLS state list. */
 284         state_lock(ha);
 285         cds_list_add(&URCU_TLS(health_state).node, &ha->list);
 286         state_unlock(ha);
 287 }
 288
 289 /*
 290  * Remove node from global list.
 291  */
 292 void health_unregister(struct health_app *ha)
 293 {
 294         state_lock(ha);
 295         /*
 296          * On error, set the global_error_state since we are about to remove
 297          * the node from the global list.
 298          */
 299         if (uatomic_read(&URCU_TLS(health_state).flags) & HEALTH_ERROR) {
 300                 uatomic_set(&ha->flags[URCU_TLS(health_state).type],
 301                                 HEALTH_ERROR);
 302         }
 303         cds_list_del(&URCU_TLS(health_state).node);
 304         state_unlock(ha);
 305 }