| 1 |
/** |
| 2 |
* collectd - src/ceph.c |
| 3 |
* Copyright (C) 2011 New Dream Network |
| 4 |
* Copyright (C) 2015 Florian octo Forster |
| 5 |
* |
| 6 |
* This program is free software; you can redistribute it and/or modify it |
| 7 |
* under the terms of the GNU General Public License as published by the |
| 8 |
* Free Software Foundation; only version 2 of the License is applicable. |
| 9 |
* |
| 10 |
* This program is distributed in the hope that it will be useful, but |
| 11 |
* WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 |
* General Public License for more details. |
| 14 |
* |
| 15 |
* You should have received a copy of the GNU General Public License along |
| 16 |
* with this program; if not, write to the Free Software Foundation, Inc., |
| 17 |
* 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 18 |
* |
| 19 |
* Authors: |
| 20 |
* Colin McCabe <cmccabe at alumni.cmu.edu> |
| 21 |
* Dennis Zou <yunzou at cisco.com> |
| 22 |
* Dan Ryder <daryder at cisco.com> |
| 23 |
* Florian octo Forster <octo at collectd.org> |
| 24 |
**/ |
| 25 |
|
| 26 |
#define _DEFAULT_SOURCE |
| 27 |
#define _BSD_SOURCE |
| 28 |
|
| 29 |
#include "collectd.h" |
| 30 |
|
| 31 |
#include "plugin.h" |
| 32 |
#include "utils/common/common.h" |
| 33 |
|
| 34 |
#include <arpa/inet.h> |
| 35 |
#include <errno.h> |
| 36 |
#include <fcntl.h> |
| 37 |
#include <yajl/yajl_parse.h> |
| 38 |
#if HAVE_YAJL_YAJL_VERSION_H |
| 39 |
#include <yajl/yajl_version.h> |
| 40 |
#endif |
| 41 |
#ifdef HAVE_SYS_CAPABILITY_H |
| 42 |
#include <sys/capability.h> |
| 43 |
#endif |
| 44 |
|
| 45 |
#include <inttypes.h> |
| 46 |
#include <limits.h> |
| 47 |
#include <math.h> |
| 48 |
#include <poll.h> |
| 49 |
#include <stdint.h> |
| 50 |
#include <stdio.h> |
| 51 |
#include <stdlib.h> |
| 52 |
#include <string.h> |
| 53 |
#include <strings.h> |
| 54 |
#include <sys/time.h> |
| 55 |
#include <sys/types.h> |
| 56 |
#include <sys/un.h> |
| 57 |
#include <unistd.h> |
| 58 |
|
| 59 |
#define RETRY_AVGCOUNT -1 |
| 60 |
|
| 61 |
#if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1) |
| 62 |
#define HAVE_YAJL_V2 1 |
| 63 |
#endif |
| 64 |
|
| 65 |
#define RETRY_ON_EINTR(ret, expr) \ |
| 66 |
while (1) { \ |
| 67 |
ret = expr; \ |
| 68 |
if (ret >= 0) \ |
| 69 |
break; \ |
| 70 |
ret = -errno; \ |
| 71 |
if (ret != -EINTR) \ |
| 72 |
break; \ |
| 73 |
} |
| 74 |
|
| 75 |
/** Timeout interval in seconds */ |
| 76 |
#define CEPH_TIMEOUT_INTERVAL 1 |
| 77 |
|
| 78 |
/** Maximum path length for a UNIX domain socket on this system */ |
| 79 |
#define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un *)0)->sun_path)) |
| 80 |
|
| 81 |
/** Yajl callback returns */ |
| 82 |
#define CEPH_CB_CONTINUE 1 |
| 83 |
#define CEPH_CB_ABORT 0 |
| 84 |
|
| 85 |
#if HAVE_YAJL_V2 |
| 86 |
typedef size_t yajl_len_t; |
| 87 |
#else |
| 88 |
typedef unsigned int yajl_len_t; |
| 89 |
#endif |
| 90 |
|
| 91 |
/** Number of types for ceph defined in types.db */ |
| 92 |
#define CEPH_DSET_TYPES_NUM 3 |
| 93 |
/** ceph types enum */ |
| 94 |
enum ceph_dset_type_d { |
| 95 |
DSET_LATENCY = 0, |
| 96 |
DSET_BYTES = 1, |
| 97 |
DSET_RATE = 2, |
| 98 |
DSET_TYPE_UNFOUND = 1000 |
| 99 |
}; |
| 100 |
|
| 101 |
/** Valid types for ceph defined in types.db */ |
| 102 |
static const char *const ceph_dset_types[CEPH_DSET_TYPES_NUM] = { |
| 103 |
"ceph_latency", "ceph_bytes", "ceph_rate"}; |
| 104 |
|
| 105 |
/******* ceph_daemon *******/ |
| 106 |
struct ceph_daemon { |
| 107 |
/** Version of the admin_socket interface */ |
| 108 |
uint32_t version; |
| 109 |
/** daemon name **/ |
| 110 |
char name[DATA_MAX_NAME_LEN]; |
| 111 |
|
| 112 |
/** Path to the socket that we use to talk to the ceph daemon */ |
| 113 |
char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX]; |
| 114 |
|
| 115 |
/** Number of counters */ |
| 116 |
int ds_num; |
| 117 |
/** Track ds types */ |
| 118 |
uint32_t *ds_types; |
| 119 |
/** Track ds names to match with types */ |
| 120 |
char **ds_names; |
| 121 |
|
| 122 |
/** |
| 123 |
* Keep track of last data for latency values so we can calculate rate |
| 124 |
* since last poll. |
| 125 |
*/ |
| 126 |
struct last_data **last_poll_data; |
| 127 |
/** index of last poll data */ |
| 128 |
int last_idx; |
| 129 |
}; |
| 130 |
|
| 131 |
/******* JSON parsing *******/ |
| 132 |
typedef int (*node_handler_t)(void *, const char *, const char *); |
| 133 |
|
| 134 |
/** Track state and handler while parsing JSON */ |
| 135 |
struct yajl_struct { |
| 136 |
node_handler_t handler; |
| 137 |
void *handler_arg; |
| 138 |
|
| 139 |
char *key; |
| 140 |
char *stack[YAJL_MAX_DEPTH]; |
| 141 |
size_t depth; |
| 142 |
}; |
| 143 |
typedef struct yajl_struct yajl_struct; |
| 144 |
|
| 145 |
enum perfcounter_type_d { |
| 146 |
PERFCOUNTER_LATENCY = 0x4, |
| 147 |
PERFCOUNTER_DERIVE = 0x8, |
| 148 |
}; |
| 149 |
|
| 150 |
/** Give user option to use default (long run = since daemon started) avg */ |
| 151 |
static int long_run_latency_avg; |
| 152 |
|
| 153 |
/** |
| 154 |
* Give user option to use default type for special cases - |
| 155 |
* filestore.journal_wr_bytes is currently only metric here. Ceph reports the |
| 156 |
* type as a sum/count pair and will calculate it the same as a latency value. |
| 157 |
* All other "bytes" metrics (excluding the used/capacity bytes for the OSD) |
| 158 |
* use the DERIVE type. Unless user specifies to use given type, convert this |
| 159 |
* metric to use DERIVE. |
| 160 |
*/ |
| 161 |
static int convert_special_metrics = 1; |
| 162 |
|
| 163 |
/** Array of daemons to monitor */ |
| 164 |
static struct ceph_daemon **g_daemons; |
| 165 |
|
| 166 |
/** Number of elements in g_daemons */ |
| 167 |
static size_t g_num_daemons; |
| 168 |
|
| 169 |
/** |
| 170 |
* A set of data that we build up in memory while parsing the JSON. |
| 171 |
*/ |
| 172 |
struct values_tmp { |
| 173 |
/** ceph daemon we are processing data for*/ |
| 174 |
struct ceph_daemon *d; |
| 175 |
/** track avgcount across counters for avgcount/sum latency pairs */ |
| 176 |
uint64_t avgcount; |
| 177 |
/** current index of counters - used to get type of counter */ |
| 178 |
int index; |
| 179 |
/** |
| 180 |
* similar to index, but current index of latency type counters - |
| 181 |
* used to get last poll data of counter |
| 182 |
*/ |
| 183 |
int latency_index; |
| 184 |
/** |
| 185 |
* values list - maintain across counters since |
| 186 |
* host/plugin/plugin instance are always the same |
| 187 |
*/ |
| 188 |
value_list_t vlist; |
| 189 |
}; |
| 190 |
|
| 191 |
/** |
| 192 |
* A set of count/sum pairs to keep track of latency types and get difference |
| 193 |
* between this poll data and last poll data. |
| 194 |
*/ |
| 195 |
struct last_data { |
| 196 |
char ds_name[DATA_MAX_NAME_LEN]; |
| 197 |
double last_sum; |
| 198 |
uint64_t last_count; |
| 199 |
}; |
| 200 |
|
| 201 |
/******* network I/O *******/ |
| 202 |
enum cstate_t { |
| 203 |
CSTATE_UNCONNECTED = 0, |
| 204 |
CSTATE_WRITE_REQUEST, |
| 205 |
CSTATE_READ_VERSION, |
| 206 |
CSTATE_READ_AMT, |
| 207 |
CSTATE_READ_JSON, |
| 208 |
}; |
| 209 |
|
| 210 |
enum request_type_t { |
| 211 |
ASOK_REQ_VERSION = 0, |
| 212 |
ASOK_REQ_DATA = 1, |
| 213 |
ASOK_REQ_SCHEMA = 2, |
| 214 |
ASOK_REQ_NONE = 1000, |
| 215 |
}; |
| 216 |
|
| 217 |
struct cconn { |
| 218 |
/** The Ceph daemon that we're talking to */ |
| 219 |
struct ceph_daemon *d; |
| 220 |
|
| 221 |
/** Request type */ |
| 222 |
uint32_t request_type; |
| 223 |
|
| 224 |
/** The connection state */ |
| 225 |
enum cstate_t state; |
| 226 |
|
| 227 |
/** The socket we use to talk to this daemon */ |
| 228 |
int asok; |
| 229 |
|
| 230 |
/** The amount of data remaining to read / write. */ |
| 231 |
uint32_t amt; |
| 232 |
|
| 233 |
/** Length of the JSON to read */ |
| 234 |
uint32_t json_len; |
| 235 |
|
| 236 |
/** Buffer containing JSON data */ |
| 237 |
unsigned char *json; |
| 238 |
|
| 239 |
/** Keep data important to yajl processing */ |
| 240 |
struct yajl_struct yajl; |
| 241 |
}; |
| 242 |
|
| 243 |
static int ceph_cb_null(void *ctx) { return CEPH_CB_CONTINUE; } |
| 244 |
|
| 245 |
static int ceph_cb_boolean(void *ctx, int bool_val) { return CEPH_CB_CONTINUE; } |
| 246 |
|
| 247 |
#define BUFFER_ADD(dest, src) \ |
| 248 |
do { \ |
| 249 |
size_t dest_size = sizeof(dest); \ |
| 250 |
size_t dest_len = strlen(dest); \ |
| 251 |
if (dest_size > dest_len) { \ |
| 252 |
sstrncpy((dest) + dest_len, (src), dest_size - dest_len); \ |
| 253 |
} \ |
| 254 |
(dest)[dest_size - 1] = '\0'; \ |
| 255 |
} while (0) |
| 256 |
|
| 257 |
static int ceph_cb_number(void *ctx, const char *number_val, |
| 258 |
yajl_len_t number_len) { |
| 259 |
yajl_struct *state = (yajl_struct *)ctx; |
| 260 |
char buffer[number_len + 1]; |
| 261 |
char key[2 * DATA_MAX_NAME_LEN] = {0}; |
| 262 |
int status; |
| 263 |
|
| 264 |
memcpy(buffer, number_val, number_len); |
| 265 |
buffer[sizeof(buffer) - 1] = '\0'; |
| 266 |
|
| 267 |
for (size_t i = 0; i < state->depth; i++) { |
| 268 |
if (state->stack[i] == NULL) |
| 269 |
continue; |
| 270 |
|
| 271 |
if (strlen(key) != 0) |
| 272 |
BUFFER_ADD(key, "."); |
| 273 |
BUFFER_ADD(key, state->stack[i]); |
| 274 |
} |
| 275 |
|
| 276 |
/* Super-special case for filestore.journal_wr_bytes.avgcount: For |
| 277 |
* some reason, Ceph schema encodes this as a count/sum pair while all |
| 278 |
* other "Bytes" data (excluding used/capacity bytes for OSD space) uses |
| 279 |
* a single "Derive" type. To spare further confusion, keep this KPI as |
| 280 |
* the same type of other "Bytes". Instead of keeping an "average" or |
| 281 |
* "rate", use the "sum" in the pair and assign that to the derive |
| 282 |
* value. */ |
| 283 |
if (convert_special_metrics && (state->depth > 2) && |
| 284 |
state->stack[state->depth - 2] && |
| 285 |
(strcmp("filestore", state->stack[state->depth - 2]) == 0) && |
| 286 |
state->stack[state->depth - 1] && |
| 287 |
(strcmp("journal_wr_bytes", state->stack[state->depth - 1]) == 0) && |
| 288 |
(strcmp("avgcount", state->key) == 0)) { |
| 289 |
DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes"); |
| 290 |
return CEPH_CB_CONTINUE; |
| 291 |
} |
| 292 |
|
| 293 |
BUFFER_ADD(key, "."); |
| 294 |
BUFFER_ADD(key, state->key); |
| 295 |
|
| 296 |
status = state->handler(state->handler_arg, buffer, key); |
| 297 |
|
| 298 |
if (status != 0) { |
| 299 |
ERROR("ceph plugin: JSON handler failed with status %d.", status); |
| 300 |
return CEPH_CB_ABORT; |
| 301 |
} |
| 302 |
|
| 303 |
return CEPH_CB_CONTINUE; |
| 304 |
} |
| 305 |
|
| 306 |
static int ceph_cb_string(void *ctx, const unsigned char *string_val, |
| 307 |
yajl_len_t string_len) { |
| 308 |
return CEPH_CB_CONTINUE; |
| 309 |
} |
| 310 |
|
| 311 |
static int ceph_cb_start_map(void *ctx) { |
| 312 |
yajl_struct *state = (yajl_struct *)ctx; |
| 313 |
|
| 314 |
/* Push key to the stack */ |
| 315 |
if (state->depth == YAJL_MAX_DEPTH) |
| 316 |
return CEPH_CB_ABORT; |
| 317 |
|
| 318 |
state->stack[state->depth] = state->key; |
| 319 |
state->depth++; |
| 320 |
state->key = NULL; |
| 321 |
|
| 322 |
return CEPH_CB_CONTINUE; |
| 323 |
} |
| 324 |
|
| 325 |
static int ceph_cb_end_map(void *ctx) { |
| 326 |
yajl_struct *state = (yajl_struct *)ctx; |
| 327 |
|
| 328 |
/* Pop key from the stack */ |
| 329 |
if (state->depth == 0) |
| 330 |
return CEPH_CB_ABORT; |
| 331 |
|
| 332 |
sfree(state->key); |
| 333 |
state->depth--; |
| 334 |
state->key = state->stack[state->depth]; |
| 335 |
state->stack[state->depth] = NULL; |
| 336 |
|
| 337 |
return CEPH_CB_CONTINUE; |
| 338 |
} |
| 339 |
|
| 340 |
static int ceph_cb_map_key(void *ctx, const unsigned char *key, |
| 341 |
yajl_len_t string_len) { |
| 342 |
yajl_struct *state = (yajl_struct *)ctx; |
| 343 |
size_t sz = ((size_t)string_len) + 1; |
| 344 |
|
| 345 |
sfree(state->key); |
| 346 |
state->key = malloc(sz); |
| 347 |
if (state->key == NULL) { |
| 348 |
ERROR("ceph plugin: malloc failed."); |
| 349 |
return CEPH_CB_ABORT; |
| 350 |
} |
| 351 |
|
| 352 |
memmove(state->key, key, sz - 1); |
| 353 |
state->key[sz - 1] = '\0'; |
| 354 |
|
| 355 |
return CEPH_CB_CONTINUE; |
| 356 |
} |
| 357 |
|
| 358 |
static int ceph_cb_start_array(void *ctx) { return CEPH_CB_CONTINUE; } |
| 359 |
|
| 360 |
static int ceph_cb_end_array(void *ctx) { return CEPH_CB_CONTINUE; } |
| 361 |
|
| 362 |
static yajl_callbacks callbacks = {ceph_cb_null, |
| 363 |
ceph_cb_boolean, |
| 364 |
NULL, |
| 365 |
NULL, |
| 366 |
ceph_cb_number, |
| 367 |
ceph_cb_string, |
| 368 |
ceph_cb_start_map, |
| 369 |
ceph_cb_map_key, |
| 370 |
ceph_cb_end_map, |
| 371 |
ceph_cb_start_array, |
| 372 |
ceph_cb_end_array}; |
| 373 |
|
| 374 |
static void ceph_daemon_print(const struct ceph_daemon *d) { |
| 375 |
DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path); |
| 376 |
} |
| 377 |
|
| 378 |
static void ceph_daemons_print(void) { |
| 379 |
for (size_t i = 0; i < g_num_daemons; ++i) { |
| 380 |
ceph_daemon_print(g_daemons[i]); |
| 381 |
} |
| 382 |
} |
| 383 |
|
| 384 |
static void ceph_daemon_free(struct ceph_daemon *d) { |
| 385 |
for (int i = 0; i < d->last_idx; i++) { |
| 386 |
sfree(d->last_poll_data[i]); |
| 387 |
} |
| 388 |
sfree(d->last_poll_data); |
| 389 |
d->last_poll_data = NULL; |
| 390 |
d->last_idx = 0; |
| 391 |
|
| 392 |
for (int i = 0; i < d->ds_num; i++) { |
| 393 |
sfree(d->ds_names[i]); |
| 394 |
} |
| 395 |
sfree(d->ds_types); |
| 396 |
sfree(d->ds_names); |
| 397 |
sfree(d); |
| 398 |
} |
| 399 |
|
| 400 |
/* compact_ds_name removed the special characters ":", "_", "-" and "+" from the |
| 401 |
* input string. Characters following these special characters are capitalized. |
| 402 |
* Trailing "+" and "-" characters are replaces with the strings "Plus" and |
| 403 |
* "Minus". */ |
| 404 |
static int compact_ds_name(char *buffer, size_t buffer_size, char const *src) { |
| 405 |
char *src_copy; |
| 406 |
size_t src_len; |
| 407 |
char *ptr = buffer; |
| 408 |
size_t ptr_size = buffer_size; |
| 409 |
bool append_plus = false; |
| 410 |
bool append_minus = false; |
| 411 |
|
| 412 |
if ((buffer == NULL) || (buffer_size <= strlen("Minus")) || (src == NULL)) |
| 413 |
return EINVAL; |
| 414 |
|
| 415 |
src_copy = strdup(src); |
| 416 |
src_len = strlen(src); |
| 417 |
|
| 418 |
/* Remove trailing "+" and "-". */ |
| 419 |
if (src_copy[src_len - 1] == '+') { |
| 420 |
append_plus = true; |
| 421 |
src_len--; |
| 422 |
src_copy[src_len] = 0; |
| 423 |
} else if (src_copy[src_len - 1] == '-') { |
| 424 |
append_minus = true; |
| 425 |
src_len--; |
| 426 |
src_copy[src_len] = 0; |
| 427 |
} |
| 428 |
|
| 429 |
/* Split at special chars, capitalize first character, append to buffer. */ |
| 430 |
char *dummy = src_copy; |
| 431 |
char *token; |
| 432 |
char *save_ptr = NULL; |
| 433 |
while ((token = strtok_r(dummy, ":_-+", &save_ptr)) != NULL) { |
| 434 |
size_t len; |
| 435 |
|
| 436 |
dummy = NULL; |
| 437 |
|
| 438 |
token[0] = toupper((int)token[0]); |
| 439 |
|
| 440 |
assert(ptr_size > 1); |
| 441 |
|
| 442 |
len = strlen(token); |
| 443 |
if (len >= ptr_size) |
| 444 |
len = ptr_size - 1; |
| 445 |
|
| 446 |
assert(len > 0); |
| 447 |
assert(len < ptr_size); |
| 448 |
|
| 449 |
sstrncpy(ptr, token, len + 1); |
| 450 |
ptr += len; |
| 451 |
ptr_size -= len; |
| 452 |
|
| 453 |
assert(*ptr == 0); |
| 454 |
if (ptr_size <= 1) |
| 455 |
break; |
| 456 |
} |
| 457 |
|
| 458 |
/* Append "Plus" or "Minus" if "+" or "-" has been stripped above. */ |
| 459 |
if (append_plus || append_minus) { |
| 460 |
char const *append = "Plus"; |
| 461 |
if (append_minus) |
| 462 |
append = "Minus"; |
| 463 |
|
| 464 |
size_t offset = buffer_size - (strlen(append) + 1); |
| 465 |
if (offset > strlen(buffer)) |
| 466 |
offset = strlen(buffer); |
| 467 |
|
| 468 |
sstrncpy(buffer + offset, append, buffer_size - offset); |
| 469 |
} |
| 470 |
|
| 471 |
sfree(src_copy); |
| 472 |
return 0; |
| 473 |
} |
| 474 |
|
| 475 |
static bool has_suffix(char const *str, char const *suffix) { |
| 476 |
size_t str_len = strlen(str); |
| 477 |
size_t suffix_len = strlen(suffix); |
| 478 |
size_t offset; |
| 479 |
|
| 480 |
if (suffix_len > str_len) |
| 481 |
return false; |
| 482 |
offset = str_len - suffix_len; |
| 483 |
|
| 484 |
if (strcmp(str + offset, suffix) == 0) |
| 485 |
return true; |
| 486 |
|
| 487 |
return false; |
| 488 |
} |
| 489 |
|
| 490 |
static void cut_suffix(char *buffer, size_t buffer_size, char const *str, |
| 491 |
char const *suffix) { |
| 492 |
|
| 493 |
size_t str_len = strlen(str); |
| 494 |
size_t suffix_len = strlen(suffix); |
| 495 |
|
| 496 |
size_t offset = str_len - suffix_len + 1; |
| 497 |
|
| 498 |
if (offset > buffer_size) { |
| 499 |
offset = buffer_size; |
| 500 |
} |
| 501 |
|
| 502 |
sstrncpy(buffer, str, offset); |
| 503 |
} |
| 504 |
|
| 505 |
/* count_parts returns the number of elements a "foo.bar.baz" style key has. */ |
| 506 |
static size_t count_parts(char const *key) { |
| 507 |
size_t parts_num = 0; |
| 508 |
|
| 509 |
for (const char *ptr = key; ptr != NULL; ptr = strchr(ptr + 1, '.')) |
| 510 |
parts_num++; |
| 511 |
|
| 512 |
return parts_num; |
| 513 |
} |
| 514 |
|
| 515 |
/** |
| 516 |
* Parse key to remove "type" if this is for schema and initiate compaction |
| 517 |
*/ |
| 518 |
static int parse_keys(char *buffer, size_t buffer_size, const char *key_str) { |
| 519 |
char tmp[2 * buffer_size]; |
| 520 |
size_t tmp_size = sizeof(tmp); |
| 521 |
const char *cut_suffixes[] = {".type", ".avgcount", ".sum", ".avgtime"}; |
| 522 |
|
| 523 |
if (buffer == NULL || buffer_size == 0 || key_str == NULL || |
| 524 |
strlen(key_str) == 0) |
| 525 |
return EINVAL; |
| 526 |
|
| 527 |
sstrncpy(tmp, key_str, tmp_size); |
| 528 |
|
| 529 |
/* Strip suffix if it is ".type" or one of latency metric suffix. */ |
| 530 |
if (count_parts(key_str) > 2) { |
| 531 |
for (size_t i = 0; i < STATIC_ARRAY_SIZE(cut_suffixes); i++) { |
| 532 |
if (has_suffix(key_str, cut_suffixes[i])) { |
| 533 |
cut_suffix(tmp, tmp_size, key_str, cut_suffixes[i]); |
| 534 |
break; |
| 535 |
} |
| 536 |
} |
| 537 |
} |
| 538 |
|
| 539 |
return compact_ds_name(buffer, buffer_size, tmp); |
| 540 |
} |
| 541 |
|
| 542 |
/** |
| 543 |
* while parsing ceph admin socket schema, save counter name and type for later |
| 544 |
* data processing |
| 545 |
*/ |
| 546 |
static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name, |
| 547 |
int pc_type) { |
| 548 |
uint32_t type; |
| 549 |
char ds_name[DATA_MAX_NAME_LEN]; |
| 550 |
|
| 551 |
if (convert_special_metrics) { |
| 552 |
/** |
| 553 |
* Special case for filestore:JournalWrBytes. For some reason, Ceph |
| 554 |
* schema encodes this as a count/sum pair while all other "Bytes" data |
| 555 |
* (excluding used/capacity bytes for OSD space) uses a single "Derive" |
| 556 |
* type. To spare further confusion, keep this KPI as the same type of |
| 557 |
* other "Bytes". Instead of keeping an "average" or "rate", use the |
| 558 |
* "sum" in the pair and assign that to the derive value. |
| 559 |
*/ |
| 560 |
if ((strcmp(name, "filestore.journal_wr_bytes.type") == 0)) { |
| 561 |
pc_type = 10; |
| 562 |
} |
| 563 |
} |
| 564 |
|
| 565 |
d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1)); |
| 566 |
if (!d->ds_names) { |
| 567 |
return -ENOMEM; |
| 568 |
} |
| 569 |
|
| 570 |
d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1)); |
| 571 |
if (!d->ds_types) { |
| 572 |
return -ENOMEM; |
| 573 |
} |
| 574 |
|
| 575 |
d->ds_names[d->ds_num] = malloc(DATA_MAX_NAME_LEN); |
| 576 |
if (!d->ds_names[d->ds_num]) { |
| 577 |
return -ENOMEM; |
| 578 |
} |
| 579 |
|
| 580 |
type = (pc_type & PERFCOUNTER_DERIVE) |
| 581 |
? DSET_RATE |
| 582 |
: ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES); |
| 583 |
d->ds_types[d->ds_num] = type; |
| 584 |
|
| 585 |
if (parse_keys(ds_name, sizeof(ds_name), name)) { |
| 586 |
return 1; |
| 587 |
} |
| 588 |
|
| 589 |
sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN - 1); |
| 590 |
d->ds_num = (d->ds_num + 1); |
| 591 |
|
| 592 |
return 0; |
| 593 |
} |
| 594 |
|
| 595 |
/******* ceph_config *******/ |
| 596 |
static int cc_handle_str(struct oconfig_item_s *item, char *dest, |
| 597 |
int dest_len) { |
| 598 |
const char *val; |
| 599 |
if (item->values_num != 1) { |
| 600 |
return -ENOTSUP; |
| 601 |
} |
| 602 |
if (item->values[0].type != OCONFIG_TYPE_STRING) { |
| 603 |
return -ENOTSUP; |
| 604 |
} |
| 605 |
val = item->values[0].value.string; |
| 606 |
if (snprintf(dest, dest_len, "%s", val) > (dest_len - 1)) { |
| 607 |
ERROR("ceph plugin: configuration parameter '%s' is too long.\n", |
| 608 |
item->key); |
| 609 |
return -ENAMETOOLONG; |
| 610 |
} |
| 611 |
return 0; |
| 612 |
} |
| 613 |
|
| 614 |
static int cc_handle_bool(struct oconfig_item_s *item, int *dest) { |
| 615 |
if (item->values_num != 1) { |
| 616 |
return -ENOTSUP; |
| 617 |
} |
| 618 |
|
| 619 |
if (item->values[0].type != OCONFIG_TYPE_BOOLEAN) { |
| 620 |
return -ENOTSUP; |
| 621 |
} |
| 622 |
|
| 623 |
*dest = (item->values[0].value.boolean) ? 1 : 0; |
| 624 |
return 0; |
| 625 |
} |
| 626 |
|
| 627 |
static int cc_add_daemon_config(oconfig_item_t *ci) { |
| 628 |
int ret; |
| 629 |
struct ceph_daemon *nd, cd = {0}; |
| 630 |
struct ceph_daemon **tmp; |
| 631 |
|
| 632 |
if ((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING)) { |
| 633 |
WARNING("ceph plugin: `Daemon' blocks need exactly one string " |
| 634 |
"argument."); |
| 635 |
return -1; |
| 636 |
} |
| 637 |
|
| 638 |
ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN); |
| 639 |
if (ret) { |
| 640 |
return ret; |
| 641 |
} |
| 642 |
|
| 643 |
for (int i = 0; i < ci->children_num; i++) { |
| 644 |
oconfig_item_t *child = ci->children + i; |
| 645 |
|
| 646 |
if (strcasecmp("SocketPath", child->key) == 0) { |
| 647 |
ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path)); |
| 648 |
if (ret) { |
| 649 |
return ret; |
| 650 |
} |
| 651 |
} else { |
| 652 |
WARNING("ceph plugin: ignoring unknown option %s", child->key); |
| 653 |
} |
| 654 |
} |
| 655 |
if (cd.name[0] == '\0') { |
| 656 |
ERROR("ceph plugin: you must configure a daemon name.\n"); |
| 657 |
return -EINVAL; |
| 658 |
} else if (cd.asok_path[0] == '\0') { |
| 659 |
ERROR("ceph plugin(name=%s): you must configure an administrative " |
| 660 |
"socket path.\n", |
| 661 |
cd.name); |
| 662 |
return -EINVAL; |
| 663 |
} else if (!((cd.asok_path[0] == '/') || |
| 664 |
(cd.asok_path[0] == '.' && cd.asok_path[1] == '/'))) { |
| 665 |
ERROR("ceph plugin(name=%s): administrative socket paths must begin " |
| 666 |
"with '/' or './' Can't parse: '%s'\n", |
| 667 |
cd.name, cd.asok_path); |
| 668 |
return -EINVAL; |
| 669 |
} |
| 670 |
|
| 671 |
tmp = realloc(g_daemons, (g_num_daemons + 1) * sizeof(*g_daemons)); |
| 672 |
if (tmp == NULL) { |
| 673 |
/* The positive return value here indicates that this is a |
| 674 |
* runtime error, not a configuration error. */ |
| 675 |
return ENOMEM; |
| 676 |
} |
| 677 |
g_daemons = tmp; |
| 678 |
|
| 679 |
nd = malloc(sizeof(*nd)); |
| 680 |
if (!nd) { |
| 681 |
return ENOMEM; |
| 682 |
} |
| 683 |
memcpy(nd, &cd, sizeof(*nd)); |
| 684 |
g_daemons[g_num_daemons] = nd; |
| 685 |
g_num_daemons++; |
| 686 |
return 0; |
| 687 |
} |
| 688 |
|
| 689 |
static int ceph_config(oconfig_item_t *ci) { |
| 690 |
int ret; |
| 691 |
|
| 692 |
for (int i = 0; i < ci->children_num; ++i) { |
| 693 |
oconfig_item_t *child = ci->children + i; |
| 694 |
if (strcasecmp("Daemon", child->key) == 0) { |
| 695 |
ret = cc_add_daemon_config(child); |
| 696 |
if (ret == ENOMEM) { |
| 697 |
ERROR("ceph plugin: Couldn't allocate memory"); |
| 698 |
return ret; |
| 699 |
} else if (ret) { |
| 700 |
// process other daemons and ignore this one |
| 701 |
continue; |
| 702 |
} |
| 703 |
} else if (strcasecmp("LongRunAvgLatency", child->key) == 0) { |
| 704 |
ret = cc_handle_bool(child, &long_run_latency_avg); |
| 705 |
if (ret) { |
| 706 |
return ret; |
| 707 |
} |
| 708 |
} else if (strcasecmp("ConvertSpecialMetricTypes", child->key) == 0) { |
| 709 |
ret = cc_handle_bool(child, &convert_special_metrics); |
| 710 |
if (ret) { |
| 711 |
return ret; |
| 712 |
} |
| 713 |
} else { |
| 714 |
WARNING("ceph plugin: ignoring unknown option %s", child->key); |
| 715 |
} |
| 716 |
} |
| 717 |
return 0; |
| 718 |
} |
| 719 |
|
| 720 |
/** |
| 721 |
* Parse JSON and get error message if present |
| 722 |
*/ |
| 723 |
static int traverse_json(const unsigned char *json, uint32_t json_len, |
| 724 |
yajl_handle hand) { |
| 725 |
yajl_status status = yajl_parse(hand, json, json_len); |
| 726 |
unsigned char *msg; |
| 727 |
|
| 728 |
switch (status) { |
| 729 |
case yajl_status_error: |
| 730 |
msg = yajl_get_error(hand, /* verbose = */ 1, |
| 731 |
/* jsonText = */ (unsigned char *)json, |
| 732 |
(unsigned int)json_len); |
| 733 |
ERROR("ceph plugin: yajl_parse failed: %s", msg); |
| 734 |
yajl_free_error(hand, msg); |
| 735 |
return 1; |
| 736 |
case yajl_status_client_canceled: |
| 737 |
return 1; |
| 738 |
default: |
| 739 |
return 0; |
| 740 |
} |
| 741 |
} |
| 742 |
|
| 743 |
/** |
| 744 |
* Add entry for each counter while parsing schema |
| 745 |
*/ |
| 746 |
static int node_handler_define_schema(void *arg, const char *val, |
| 747 |
const char *key) { |
| 748 |
struct ceph_daemon *d = (struct ceph_daemon *)arg; |
| 749 |
int pc_type; |
| 750 |
pc_type = atoi(val); |
| 751 |
return ceph_daemon_add_ds_entry(d, key, pc_type); |
| 752 |
} |
| 753 |
|
| 754 |
/** |
| 755 |
* Latency counter does not yet have an entry in last poll data - add it. |
| 756 |
*/ |
| 757 |
static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum, |
| 758 |
uint64_t cur_count) { |
| 759 |
d->last_poll_data[d->last_idx] = |
| 760 |
malloc(sizeof(*d->last_poll_data[d->last_idx])); |
| 761 |
if (!d->last_poll_data[d->last_idx]) { |
| 762 |
return -ENOMEM; |
| 763 |
} |
| 764 |
sstrncpy(d->last_poll_data[d->last_idx]->ds_name, ds_n, |
| 765 |
sizeof(d->last_poll_data[d->last_idx]->ds_name)); |
| 766 |
d->last_poll_data[d->last_idx]->last_sum = cur_sum; |
| 767 |
d->last_poll_data[d->last_idx]->last_count = cur_count; |
| 768 |
d->last_idx = (d->last_idx + 1); |
| 769 |
return 0; |
| 770 |
} |
| 771 |
|
| 772 |
/** |
| 773 |
* Update latency counter or add new entry if it doesn't exist |
| 774 |
*/ |
| 775 |
static int update_last(struct ceph_daemon *d, const char *ds_n, int index, |
| 776 |
double cur_sum, uint64_t cur_count) { |
| 777 |
if ((d->last_idx > index) && |
| 778 |
(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)) { |
| 779 |
d->last_poll_data[index]->last_sum = cur_sum; |
| 780 |
d->last_poll_data[index]->last_count = cur_count; |
| 781 |
return 0; |
| 782 |
} |
| 783 |
|
| 784 |
if (!d->last_poll_data) { |
| 785 |
d->last_poll_data = malloc(sizeof(*d->last_poll_data)); |
| 786 |
if (!d->last_poll_data) { |
| 787 |
return -ENOMEM; |
| 788 |
} |
| 789 |
} else { |
| 790 |
struct last_data **tmp_last = realloc( |
| 791 |
d->last_poll_data, ((d->last_idx + 1) * sizeof(struct last_data *))); |
| 792 |
if (!tmp_last) { |
| 793 |
return -ENOMEM; |
| 794 |
} |
| 795 |
d->last_poll_data = tmp_last; |
| 796 |
} |
| 797 |
return add_last(d, ds_n, cur_sum, cur_count); |
| 798 |
} |
| 799 |
|
| 800 |
/** |
| 801 |
* If using index guess failed (shouldn't happen, but possible if counters |
| 802 |
* get rearranged), resort to searching for counter name |
| 803 |
*/ |
| 804 |
static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n) { |
| 805 |
for (int i = 0; i < d->last_idx; i++) { |
| 806 |
if (strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0) { |
| 807 |
return i; |
| 808 |
} |
| 809 |
} |
| 810 |
return -1; |
| 811 |
} |
| 812 |
|
| 813 |
/** |
| 814 |
* Calculate average b/t current data and last poll data |
| 815 |
* if last poll data exists |
| 816 |
*/ |
| 817 |
static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index, |
| 818 |
double cur_sum, uint64_t cur_count) { |
| 819 |
double result = -1.1, sum_delt = 0.0; |
| 820 |
uint64_t count_delt = 0; |
| 821 |
int tmp_index = 0; |
| 822 |
if (d->last_idx > index) { |
| 823 |
if (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0) { |
| 824 |
tmp_index = index; |
| 825 |
} |
| 826 |
// test previous index |
| 827 |
else if ((index > 0) && |
| 828 |
(strcmp(d->last_poll_data[index - 1]->ds_name, ds_n) == 0)) { |
| 829 |
tmp_index = (index - 1); |
| 830 |
} else { |
| 831 |
tmp_index = backup_search_for_last_avg(d, ds_n); |
| 832 |
} |
| 833 |
|
| 834 |
if ((tmp_index > -1) && |
| 835 |
(cur_count > d->last_poll_data[tmp_index]->last_count)) { |
| 836 |
sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum); |
| 837 |
count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count); |
| 838 |
result = (sum_delt / count_delt); |
| 839 |
} |
| 840 |
} |
| 841 |
|
| 842 |
if (result == -1.1) { |
| 843 |
result = NAN; |
| 844 |
} |
| 845 |
if (update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM) { |
| 846 |
return -ENOMEM; |
| 847 |
} |
| 848 |
return result; |
| 849 |
} |
| 850 |
|
| 851 |
/** |
| 852 |
* If using index guess failed, resort to searching for counter name |
| 853 |
*/ |
| 854 |
static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name) { |
| 855 |
for (int i = 0; i < d->ds_num; i++) { |
| 856 |
if (strcmp(d->ds_names[i], ds_name) == 0) { |
| 857 |
return d->ds_types[i]; |
| 858 |
} |
| 859 |
} |
| 860 |
return DSET_TYPE_UNFOUND; |
| 861 |
} |
| 862 |
|
| 863 |
/** |
| 864 |
* Process counter data and dispatch values |
| 865 |
*/ |
| 866 |
static int node_handler_fetch_data(void *arg, const char *val, |
| 867 |
const char *key) { |
| 868 |
value_t uv; |
| 869 |
double tmp_d; |
| 870 |
uint64_t tmp_u; |
| 871 |
struct values_tmp *vtmp = (struct values_tmp *)arg; |
| 872 |
uint32_t type = DSET_TYPE_UNFOUND; |
| 873 |
int index = vtmp->index; |
| 874 |
|
| 875 |
char ds_name[DATA_MAX_NAME_LEN]; |
| 876 |
|
| 877 |
if (parse_keys(ds_name, sizeof(ds_name), key)) { |
| 878 |
return 1; |
| 879 |
} |
| 880 |
|
| 881 |
if (index >= vtmp->d->ds_num) { |
| 882 |
// don't overflow bounds of array |
| 883 |
index = (vtmp->d->ds_num - 1); |
| 884 |
} |
| 885 |
|
| 886 |
/** |
| 887 |
* counters should remain in same order we parsed schema... we maintain the |
| 888 |
* index variable to keep track of current point in list of counters. first |
| 889 |
* use index to guess point in array for retrieving type. if that doesn't |
| 890 |
* work, use the old way to get the counter type |
| 891 |
*/ |
| 892 |
if (strcmp(ds_name, vtmp->d->ds_names[index]) == 0) { |
| 893 |
// found match |
| 894 |
type = vtmp->d->ds_types[index]; |
| 895 |
} else if ((index > 0) && |
| 896 |
(strcmp(ds_name, vtmp->d->ds_names[index - 1]) == 0)) { |
| 897 |
// try previous key |
| 898 |
type = vtmp->d->ds_types[index - 1]; |
| 899 |
} |
| 900 |
|
| 901 |
if (type == DSET_TYPE_UNFOUND) { |
| 902 |
// couldn't find right type by guessing, check the old way |
| 903 |
type = backup_search_for_type(vtmp->d, ds_name); |
| 904 |
} |
| 905 |
|
| 906 |
switch (type) { |
| 907 |
case DSET_LATENCY: |
| 908 |
if (has_suffix(key, ".avgcount")) { |
| 909 |
sscanf(val, "%" PRIu64, &vtmp->avgcount); |
| 910 |
// return after saving avgcount - don't dispatch value |
| 911 |
// until latency calculation |
| 912 |
return 0; |
| 913 |
} else if (has_suffix(key, ".sum")) { |
| 914 |
if (vtmp->avgcount == 0) { |
| 915 |
vtmp->avgcount = 1; |
| 916 |
} |
| 917 |
// user wants latency values as long run avg |
| 918 |
// skip this step |
| 919 |
if (long_run_latency_avg) { |
| 920 |
return 0; |
| 921 |
} |
| 922 |
double sum, result; |
| 923 |
sscanf(val, "%lf", &sum); |
| 924 |
result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, |
| 925 |
vtmp->avgcount); |
| 926 |
if (result == -ENOMEM) { |
| 927 |
return -ENOMEM; |
| 928 |
} |
| 929 |
uv.gauge = result; |
| 930 |
vtmp->latency_index = (vtmp->latency_index + 1); |
| 931 |
} else if (has_suffix(key, ".avgtime")) { |
| 932 |
|
| 933 |
/* The "avgtime" metric reports ("sum" / "avgcount"), i.e. the average |
| 934 |
* time per request since the start of the Ceph daemon. Report this only |
| 935 |
* when the user has configured "long running average". Otherwise, use the |
| 936 |
* rate of "sum" and "avgcount" to calculate the current latency. |
| 937 |
*/ |
| 938 |
|
| 939 |
if (!long_run_latency_avg) { |
| 940 |
return 0; |
| 941 |
} |
| 942 |
double result; |
| 943 |
sscanf(val, "%lf", &result); |
| 944 |
uv.gauge = result; |
| 945 |
vtmp->latency_index = (vtmp->latency_index + 1); |
| 946 |
} else { |
| 947 |
WARNING("ceph plugin: ignoring unknown latency metric: %s", key); |
| 948 |
return 0; |
| 949 |
} |
| 950 |
break; |
| 951 |
case DSET_BYTES: |
| 952 |
sscanf(val, "%lf", &tmp_d); |
| 953 |
uv.gauge = tmp_d; |
| 954 |
break; |
| 955 |
case DSET_RATE: |
| 956 |
sscanf(val, "%" PRIu64, &tmp_u); |
| 957 |
uv.derive = tmp_u; |
| 958 |
break; |
| 959 |
case DSET_TYPE_UNFOUND: |
| 960 |
default: |
| 961 |
ERROR("ceph plugin: ds %s was not properly initialized.", ds_name); |
| 962 |
return -1; |
| 963 |
} |
| 964 |
|
| 965 |
sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type)); |
| 966 |
sstrncpy(vtmp->vlist.type_instance, ds_name, |
| 967 |
sizeof(vtmp->vlist.type_instance)); |
| 968 |
vtmp->vlist.values = &uv; |
| 969 |
vtmp->vlist.values_len = 1; |
| 970 |
|
| 971 |
vtmp->index = (vtmp->index + 1); |
| 972 |
plugin_dispatch_values(&vtmp->vlist); |
| 973 |
|
| 974 |
return 0; |
| 975 |
} |
| 976 |
|
| 977 |
static int cconn_connect(struct cconn *io) { |
| 978 |
struct sockaddr_un address = {0}; |
| 979 |
int flags, fd, err; |
| 980 |
if (io->state != CSTATE_UNCONNECTED) { |
| 981 |
ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED"); |
| 982 |
return -EDOM; |
| 983 |
} |
| 984 |
fd = socket(PF_UNIX, SOCK_STREAM, 0); |
| 985 |
if (fd < 0) { |
| 986 |
err = -errno; |
| 987 |
ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) " |
| 988 |
"failed: error %d", |
| 989 |
err); |
| 990 |
return err; |
| 991 |
} |
| 992 |
address.sun_family = AF_UNIX; |
| 993 |
ssnprintf(address.sun_path, sizeof(address.sun_path), "%s", io->d->asok_path); |
| 994 |
RETRY_ON_EINTR(err, connect(fd, (struct sockaddr *)&address, |
| 995 |
sizeof(struct sockaddr_un))); |
| 996 |
if (err < 0) { |
| 997 |
ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d", fd, err); |
| 998 |
close(fd); |
| 999 |
return err; |
| 1000 |
} |
| 1001 |
|
| 1002 |
flags = fcntl(fd, F_GETFL, 0); |
| 1003 |
if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0) { |
| 1004 |
err = -errno; |
| 1005 |
ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d", fd, |
| 1006 |
err); |
| 1007 |
close(fd); |
| 1008 |
return err; |
| 1009 |
} |
| 1010 |
io->asok = fd; |
| 1011 |
io->state = CSTATE_WRITE_REQUEST; |
| 1012 |
io->amt = 0; |
| 1013 |
io->json_len = 0; |
| 1014 |
io->json = NULL; |
| 1015 |
return 0; |
| 1016 |
} |
| 1017 |
|
| 1018 |
static void cconn_close(struct cconn *io) { |
| 1019 |
io->state = CSTATE_UNCONNECTED; |
| 1020 |
if (io->asok != -1) { |
| 1021 |
int res; |
| 1022 |
RETRY_ON_EINTR(res, close(io->asok)); |
| 1023 |
} |
| 1024 |
io->asok = -1; |
| 1025 |
io->amt = 0; |
| 1026 |
io->json_len = 0; |
| 1027 |
sfree(io->json); |
| 1028 |
io->json = NULL; |
| 1029 |
} |
| 1030 |
|
| 1031 |
/* Process incoming JSON counter data */ |
| 1032 |
static int cconn_process_data(struct cconn *io, yajl_struct *yajl, |
| 1033 |
yajl_handle hand) { |
| 1034 |
int ret; |
| 1035 |
struct values_tmp *vtmp = calloc(1, sizeof(*vtmp)); |
| 1036 |
if (!vtmp) { |
| 1037 |
return -ENOMEM; |
| 1038 |
} |
| 1039 |
|
| 1040 |
vtmp->vlist = (value_list_t)VALUE_LIST_INIT; |
| 1041 |
sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin)); |
| 1042 |
sstrncpy(vtmp->vlist.plugin_instance, io->d->name, |
| 1043 |
sizeof(vtmp->vlist.plugin_instance)); |
| 1044 |
|
| 1045 |
vtmp->d = io->d; |
| 1046 |
vtmp->latency_index = 0; |
| 1047 |
vtmp->index = 0; |
| 1048 |
yajl->handler_arg = vtmp; |
| 1049 |
ret = traverse_json(io->json, io->json_len, hand); |
| 1050 |
sfree(vtmp); |
| 1051 |
return ret; |
| 1052 |
} |
| 1053 |
|
| 1054 |
/** |
| 1055 |
* Initiate JSON parsing and print error if one occurs |
| 1056 |
*/ |
| 1057 |
static int cconn_process_json(struct cconn *io) { |
| 1058 |
if ((io->request_type != ASOK_REQ_DATA) && |
| 1059 |
(io->request_type != ASOK_REQ_SCHEMA)) { |
| 1060 |
return -EDOM; |
| 1061 |
} |
| 1062 |
|
| 1063 |
int result = 1; |
| 1064 |
yajl_handle hand; |
| 1065 |
yajl_status status; |
| 1066 |
|
| 1067 |
hand = yajl_alloc(&callbacks, |
| 1068 |
#if HAVE_YAJL_V2 |
| 1069 |
/* alloc funcs = */ NULL, |
| 1070 |
#else |
| 1071 |
/* alloc funcs = */ NULL, NULL, |
| 1072 |
#endif |
| 1073 |
/* context = */ (void *)(&io->yajl)); |
| 1074 |
|
| 1075 |
if (!hand) { |
| 1076 |
ERROR("ceph plugin: yajl_alloc failed."); |
| 1077 |
return ENOMEM; |
| 1078 |
} |
| 1079 |
|
| 1080 |
io->yajl.depth = 0; |
| 1081 |
|
| 1082 |
switch (io->request_type) { |
| 1083 |
case ASOK_REQ_DATA: |
| 1084 |
io->yajl.handler = node_handler_fetch_data; |
| 1085 |
result = cconn_process_data(io, &io->yajl, hand); |
| 1086 |
break; |
| 1087 |
case ASOK_REQ_SCHEMA: |
| 1088 |
// init daemon specific variables |
| 1089 |
io->d->ds_num = 0; |
| 1090 |
io->d->last_idx = 0; |
| 1091 |
io->d->last_poll_data = NULL; |
| 1092 |
io->yajl.handler = node_handler_define_schema; |
| 1093 |
io->yajl.handler_arg = io->d; |
| 1094 |
result = traverse_json(io->json, io->json_len, hand); |
| 1095 |
break; |
| 1096 |
} |
| 1097 |
|
| 1098 |
if (result) { |
| 1099 |
goto done; |
| 1100 |
} |
| 1101 |
|
| 1102 |
#if HAVE_YAJL_V2 |
| 1103 |
status = yajl_complete_parse(hand); |
| 1104 |
#else |
| 1105 |
status = yajl_parse_complete(hand); |
| 1106 |
#endif |
| 1107 |
|
| 1108 |
if (status != yajl_status_ok) { |
| 1109 |
unsigned char *errmsg = |
| 1110 |
yajl_get_error(hand, /* verbose = */ 0, |
| 1111 |
/* jsonText = */ NULL, /* jsonTextLen = */ 0); |
| 1112 |
ERROR("ceph plugin: yajl_parse_complete failed: %s", (char *)errmsg); |
| 1113 |
yajl_free_error(hand, errmsg); |
| 1114 |
yajl_free(hand); |
| 1115 |
return 1; |
| 1116 |
} |
| 1117 |
|
| 1118 |
done: |
| 1119 |
yajl_free(hand); |
| 1120 |
return result; |
| 1121 |
} |
| 1122 |
|
| 1123 |
static int cconn_validate_revents(struct cconn *io, int revents) { |
| 1124 |
if (revents & POLLERR) { |
| 1125 |
ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR", |
| 1126 |
io->d->name); |
| 1127 |
return -EIO; |
| 1128 |
} |
| 1129 |
switch (io->state) { |
| 1130 |
case CSTATE_WRITE_REQUEST: |
| 1131 |
return (revents & POLLOUT) ? 0 : -EINVAL; |
| 1132 |
case CSTATE_READ_VERSION: |
| 1133 |
case CSTATE_READ_AMT: |
| 1134 |
case CSTATE_READ_JSON: |
| 1135 |
return (revents & POLLIN) ? 0 : -EINVAL; |
| 1136 |
default: |
| 1137 |
ERROR("ceph plugin: cconn_validate_revents(name=%s) got to " |
| 1138 |
"illegal state on line %d", |
| 1139 |
io->d->name, __LINE__); |
| 1140 |
return -EDOM; |
| 1141 |
} |
| 1142 |
} |
| 1143 |
|
| 1144 |
/** Handle a network event for a connection */ |
| 1145 |
static ssize_t cconn_handle_event(struct cconn *io) { |
| 1146 |
ssize_t ret; |
| 1147 |
switch (io->state) { |
| 1148 |
case CSTATE_UNCONNECTED: |
| 1149 |
ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal " |
| 1150 |
"state on line %d", |
| 1151 |
io->d->name, __LINE__); |
| 1152 |
|
| 1153 |
return -EDOM; |
| 1154 |
case CSTATE_WRITE_REQUEST: { |
| 1155 |
char cmd[32]; |
| 1156 |
ssnprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"", io->request_type, |
| 1157 |
"\" }\n"); |
| 1158 |
size_t cmd_len = strlen(cmd); |
| 1159 |
RETRY_ON_EINTR( |
| 1160 |
ret, write(io->asok, ((char *)&cmd) + io->amt, cmd_len - io->amt)); |
| 1161 |
DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%zd)", |
| 1162 |
io->d->name, io->state, io->amt, ret); |
| 1163 |
if (ret < 0) { |
| 1164 |
return ret; |
| 1165 |
} |
| 1166 |
io->amt += ret; |
| 1167 |
if (io->amt >= cmd_len) { |
| 1168 |
io->amt = 0; |
| 1169 |
switch (io->request_type) { |
| 1170 |
case ASOK_REQ_VERSION: |
| 1171 |
io->state = CSTATE_READ_VERSION; |
| 1172 |
break; |
| 1173 |
default: |
| 1174 |
io->state = CSTATE_READ_AMT; |
| 1175 |
break; |
| 1176 |
} |
| 1177 |
} |
| 1178 |
return 0; |
| 1179 |
} |
| 1180 |
case CSTATE_READ_VERSION: { |
| 1181 |
RETRY_ON_EINTR(ret, read(io->asok, ((char *)(&io->d->version)) + io->amt, |
| 1182 |
sizeof(io->d->version) - io->amt)); |
| 1183 |
DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%zd)", |
| 1184 |
io->d->name, io->state, ret); |
| 1185 |
if (ret < 0) { |
| 1186 |
return ret; |
| 1187 |
} |
| 1188 |
io->amt += ret; |
| 1189 |
if (io->amt >= sizeof(io->d->version)) { |
| 1190 |
io->d->version = ntohl(io->d->version); |
| 1191 |
if (io->d->version != 1) { |
| 1192 |
ERROR("ceph plugin: cconn_handle_event(name=%s) not " |
| 1193 |
"expecting version %d!", |
| 1194 |
io->d->name, io->d->version); |
| 1195 |
return -ENOTSUP; |
| 1196 |
} |
| 1197 |
DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as " |
| 1198 |
"version %d", |
| 1199 |
io->d->name, io->d->version); |
| 1200 |
io->amt = 0; |
| 1201 |
cconn_close(io); |
| 1202 |
io->request_type = ASOK_REQ_SCHEMA; |
| 1203 |
} |
| 1204 |
return 0; |
| 1205 |
} |
| 1206 |
case CSTATE_READ_AMT: { |
| 1207 |
RETRY_ON_EINTR(ret, read(io->asok, ((char *)(&io->json_len)) + io->amt, |
| 1208 |
sizeof(io->json_len) - io->amt)); |
| 1209 |
DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%zd)", |
| 1210 |
io->d->name, io->state, ret); |
| 1211 |
if (ret < 0) { |
| 1212 |
return ret; |
| 1213 |
} |
| 1214 |
io->amt += ret; |
| 1215 |
if (io->amt >= sizeof(io->json_len)) { |
| 1216 |
io->json_len = ntohl(io->json_len); |
| 1217 |
io->amt = 0; |
| 1218 |
io->state = CSTATE_READ_JSON; |
| 1219 |
io->json = calloc(1, io->json_len + 1); |
| 1220 |
if (!io->json) { |
| 1221 |
ERROR("ceph plugin: error callocing io->json"); |
| 1222 |
return -ENOMEM; |
| 1223 |
} |
| 1224 |
} |
| 1225 |
return 0; |
| 1226 |
} |
| 1227 |
case CSTATE_READ_JSON: { |
| 1228 |
RETRY_ON_EINTR(ret, |
| 1229 |
read(io->asok, io->json + io->amt, io->json_len - io->amt)); |
| 1230 |
DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%zd)", |
| 1231 |
io->d->name, io->state, ret); |
| 1232 |
if (ret < 0) { |
| 1233 |
return ret; |
| 1234 |
} |
| 1235 |
io->amt += ret; |
| 1236 |
if (io->amt >= io->json_len) { |
| 1237 |
ret = cconn_process_json(io); |
| 1238 |
if (ret) { |
| 1239 |
return ret; |
| 1240 |
} |
| 1241 |
cconn_close(io); |
| 1242 |
io->request_type = ASOK_REQ_NONE; |
| 1243 |
} |
| 1244 |
return 0; |
| 1245 |
} |
| 1246 |
default: |
| 1247 |
ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal " |
| 1248 |
"state on line %d", |
| 1249 |
io->d->name, __LINE__); |
| 1250 |
return -EDOM; |
| 1251 |
} |
| 1252 |
} |
| 1253 |
|
| 1254 |
static int cconn_prepare(struct cconn *io, struct pollfd *fds) { |
| 1255 |
int ret; |
| 1256 |
if (io->request_type == ASOK_REQ_NONE) { |
| 1257 |
/* The request has already been serviced. */ |
| 1258 |
return 0; |
| 1259 |
} else if ((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0)) { |
| 1260 |
/* If there are no counters to report on, don't bother |
| 1261 |
* connecting */ |
| 1262 |
return 0; |
| 1263 |
} |
| 1264 |
|
| 1265 |
switch (io->state) { |
| 1266 |
case CSTATE_UNCONNECTED: |
| 1267 |
ret = cconn_connect(io); |
| 1268 |
if (ret > 0) { |
| 1269 |
return -ret; |
| 1270 |
} else if (ret < 0) { |
| 1271 |
return ret; |
| 1272 |
} |
| 1273 |
fds->fd = io->asok; |
| 1274 |
fds->events = POLLOUT; |
| 1275 |
return 1; |
| 1276 |
case CSTATE_WRITE_REQUEST: |
| 1277 |
fds->fd = io->asok; |
| 1278 |
fds->events = POLLOUT; |
| 1279 |
return 1; |
| 1280 |
case CSTATE_READ_VERSION: |
| 1281 |
case CSTATE_READ_AMT: |
| 1282 |
case CSTATE_READ_JSON: |
| 1283 |
fds->fd = io->asok; |
| 1284 |
fds->events = POLLIN; |
| 1285 |
return 1; |
| 1286 |
default: |
| 1287 |
ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state " |
| 1288 |
"on line %d", |
| 1289 |
io->d->name, __LINE__); |
| 1290 |
return -EDOM; |
| 1291 |
} |
| 1292 |
} |
| 1293 |
|
| 1294 |
/** Returns the difference between two struct timevals in milliseconds. |
| 1295 |
* On overflow, we return max/min int. |
| 1296 |
*/ |
| 1297 |
static int milli_diff(const struct timeval *t1, const struct timeval *t2) { |
| 1298 |
int64_t ret; |
| 1299 |
long sec_diff = t1->tv_sec - t2->tv_sec; |
| 1300 |
long usec_diff = t1->tv_usec - t2->tv_usec; |
| 1301 |
ret = usec_diff / 1000; |
| 1302 |
ret += (sec_diff * 1000); |
| 1303 |
return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret); |
| 1304 |
} |
| 1305 |
|
| 1306 |
/** This handles the actual network I/O to talk to the Ceph daemons. |
| 1307 |
*/ |
| 1308 |
static ssize_t cconn_main_loop(uint32_t request_type) { |
| 1309 |
int some_unreachable = 0; |
| 1310 |
ssize_t ret; |
| 1311 |
struct timeval end_tv; |
| 1312 |
struct cconn io_array[g_num_daemons]; |
| 1313 |
|
| 1314 |
DEBUG("ceph plugin: entering cconn_main_loop(request_type = %" PRIu32 ")", |
| 1315 |
request_type); |
| 1316 |
|
| 1317 |
if (g_num_daemons < 1) { |
| 1318 |
ERROR("ceph plugin: No daemons configured. See the \"Daemon\" config " |
| 1319 |
"option."); |
| 1320 |
return ENOENT; |
| 1321 |
} |
| 1322 |
|
| 1323 |
/* create cconn array */ |
| 1324 |
for (size_t i = 0; i < g_num_daemons; i++) { |
| 1325 |
io_array[i] = (struct cconn){ |
| 1326 |
.d = g_daemons[i], |
| 1327 |
.request_type = request_type, |
| 1328 |
.state = CSTATE_UNCONNECTED, |
| 1329 |
.asok = -1, |
| 1330 |
}; |
| 1331 |
} |
| 1332 |
|
| 1333 |
/** Calculate the time at which we should give up */ |
| 1334 |
gettimeofday(&end_tv, NULL); |
| 1335 |
end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL; |
| 1336 |
|
| 1337 |
while (1) { |
| 1338 |
int nfds, diff; |
| 1339 |
struct timeval tv; |
| 1340 |
struct cconn *polled_io_array[g_num_daemons]; |
| 1341 |
struct pollfd fds[g_num_daemons]; |
| 1342 |
memset(fds, 0, sizeof(fds)); |
| 1343 |
nfds = 0; |
| 1344 |
for (size_t i = 0; i < g_num_daemons; ++i) { |
| 1345 |
struct cconn *io = io_array + i; |
| 1346 |
ret = cconn_prepare(io, fds + nfds); |
| 1347 |
if (ret < 0) { |
| 1348 |
WARNING("ceph plugin: cconn_prepare(name=%s,i=%" PRIsz ",st=%d)=%zd", |
| 1349 |
io->d->name, i, io->state, ret); |
| 1350 |
cconn_close(io); |
| 1351 |
io->request_type = ASOK_REQ_NONE; |
| 1352 |
some_unreachable = 1; |
| 1353 |
} else if (ret == 1) { |
| 1354 |
polled_io_array[nfds++] = io_array + i; |
| 1355 |
} |
| 1356 |
} |
| 1357 |
if (nfds == 0) { |
| 1358 |
/* finished */ |
| 1359 |
ret = 0; |
| 1360 |
goto done; |
| 1361 |
} |
| 1362 |
gettimeofday(&tv, NULL); |
| 1363 |
diff = milli_diff(&end_tv, &tv); |
| 1364 |
if (diff <= 0) { |
| 1365 |
/* Timed out */ |
| 1366 |
ret = -ETIMEDOUT; |
| 1367 |
WARNING("ceph plugin: cconn_main_loop: timed out."); |
| 1368 |
goto done; |
| 1369 |
} |
| 1370 |
RETRY_ON_EINTR(ret, poll(fds, nfds, diff)); |
| 1371 |
if (ret < 0) { |
| 1372 |
ERROR("ceph plugin: poll(2) error: %zd", ret); |
| 1373 |
goto done; |
| 1374 |
} |
| 1375 |
for (int i = 0; i < nfds; ++i) { |
| 1376 |
struct cconn *io = polled_io_array[i]; |
| 1377 |
int revents = fds[i].revents; |
| 1378 |
if (revents == 0) { |
| 1379 |
/* do nothing */ |
| 1380 |
continue; |
| 1381 |
} else if (cconn_validate_revents(io, revents)) { |
| 1382 |
WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): " |
| 1383 |
"revents validation error: " |
| 1384 |
"revents=0x%08x", |
| 1385 |
io->d->name, i, io->state, revents); |
| 1386 |
cconn_close(io); |
| 1387 |
io->request_type = ASOK_REQ_NONE; |
| 1388 |
some_unreachable = 1; |
| 1389 |
} else { |
| 1390 |
ret = cconn_handle_event(io); |
| 1391 |
if (ret) { |
| 1392 |
WARNING("ceph plugin: cconn_handle_event(name=%s," |
| 1393 |
"i=%d,st=%d): error %zd", |
| 1394 |
io->d->name, i, io->state, ret); |
| 1395 |
cconn_close(io); |
| 1396 |
io->request_type = ASOK_REQ_NONE; |
| 1397 |
some_unreachable = 1; |
| 1398 |
} |
| 1399 |
} |
| 1400 |
} |
| 1401 |
} |
| 1402 |
done: |
| 1403 |
for (size_t i = 0; i < g_num_daemons; ++i) { |
| 1404 |
cconn_close(io_array + i); |
| 1405 |
} |
| 1406 |
if (some_unreachable) { |
| 1407 |
DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable."); |
| 1408 |
} else { |
| 1409 |
DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)"); |
| 1410 |
} |
| 1411 |
return ret; |
| 1412 |
} |
| 1413 |
|
| 1414 |
static int ceph_read(void) { return (int)cconn_main_loop(ASOK_REQ_DATA); } |
| 1415 |
|
| 1416 |
/******* lifecycle *******/ |
| 1417 |
static int ceph_init(void) { |
| 1418 |
#if defined(HAVE_SYS_CAPABILITY_H) && defined(CAP_DAC_OVERRIDE) |
| 1419 |
if (check_capability(CAP_DAC_OVERRIDE) != 0) { |
| 1420 |
if (getuid() == 0) |
| 1421 |
WARNING("ceph plugin: Running collectd as root, but the " |
| 1422 |
"CAP_DAC_OVERRIDE capability is missing. The plugin's read " |
| 1423 |
"function will probably fail. Is your init system dropping " |
| 1424 |
"capabilities?"); |
| 1425 |
else |
| 1426 |
WARNING( |
| 1427 |
"ceph plugin: collectd doesn't have the CAP_DAC_OVERRIDE " |
| 1428 |
"capability. If you don't want to run collectd as root, try running " |
| 1429 |
"\"setcap cap_dac_override=ep\" on the collectd binary."); |
| 1430 |
} |
| 1431 |
#endif |
| 1432 |
|
| 1433 |
ceph_daemons_print(); |
| 1434 |
|
| 1435 |
if (g_num_daemons < 1) { |
| 1436 |
ERROR("ceph plugin: No daemons configured. See the \"Daemon\" config " |
| 1437 |
"option."); |
| 1438 |
return ENOENT; |
| 1439 |
} |
| 1440 |
|
| 1441 |
return (int)cconn_main_loop(ASOK_REQ_VERSION); |
| 1442 |
} |
| 1443 |
|
| 1444 |
static int ceph_shutdown(void) { |
| 1445 |
for (size_t i = 0; i < g_num_daemons; ++i) { |
| 1446 |
ceph_daemon_free(g_daemons[i]); |
| 1447 |
} |
| 1448 |
sfree(g_daemons); |
| 1449 |
g_daemons = NULL; |
| 1450 |
g_num_daemons = 0; |
| 1451 |
DEBUG("ceph plugin: finished ceph_shutdown"); |
| 1452 |
return 0; |
| 1453 |
} |
| 1454 |
|
| 1455 |
void module_register(void) { |
| 1456 |
plugin_register_complex_config("ceph", ceph_config); |
| 1457 |
plugin_register_init("ceph", ceph_init); |
| 1458 |
plugin_register_read("ceph", ceph_read); |
| 1459 |
plugin_register_shutdown("ceph", ceph_shutdown); |
| 1460 |
} |