2 * collectd - src/ceph.c
3 * Copyright (C) 2011 New Dream Network
4 * Copyright (C) 2015 Florian octo Forster
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.
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.
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
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>
26 #define _DEFAULT_SOURCE
33 #include <arpa/inet.h>
36 #include <yajl/yajl_parse.h>
37 #if HAVE_YAJL_YAJL_VERSION_H
38 #include <yajl/yajl_version.h>
49 #include <sys/types.h>
55 #define RETRY_AVGCOUNT -1
57 #if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1)
58 # define HAVE_YAJL_V2 1
61 #define RETRY_ON_EINTR(ret, expr) \
71 /** Timeout interval in seconds */
72 #define CEPH_TIMEOUT_INTERVAL 1
74 /** Maximum path length for a UNIX domain socket on this system */
75 #define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un*)0)->sun_path))
77 /** Yajl callback returns */
78 #define CEPH_CB_CONTINUE 1
79 #define CEPH_CB_ABORT 0
82 typedef size_t yajl_len_t;
84 typedef unsigned int yajl_len_t;
87 /** Number of types for ceph defined in types.db */
88 #define CEPH_DSET_TYPES_NUM 3
89 /** ceph types enum */
95 DSET_TYPE_UNFOUND = 1000
98 /** Valid types for ceph defined in types.db */
99 static const char * const ceph_dset_types [CEPH_DSET_TYPES_NUM] =
100 {"ceph_latency", "ceph_bytes", "ceph_rate"};
102 /******* ceph_daemon *******/
105 /** Version of the admin_socket interface */
108 char name[DATA_MAX_NAME_LEN];
110 /** Path to the socket that we use to talk to the ceph daemon */
111 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
113 /** Number of counters */
115 /** Track ds types */
117 /** Track ds names to match with types */
121 * Keep track of last data for latency values so we can calculate rate
124 struct last_data **last_poll_data;
125 /** index of last poll data */
129 /******* JSON parsing *******/
130 typedef int (*node_handler_t)(void *, const char*, const char*);
132 /** Track state and handler while parsing JSON */
135 node_handler_t handler;
139 char *stack[YAJL_MAX_DEPTH];
142 typedef struct yajl_struct yajl_struct;
144 enum perfcounter_type_d
146 PERFCOUNTER_LATENCY = 0x4, PERFCOUNTER_DERIVE = 0x8,
149 /** Give user option to use default (long run = since daemon started) avg */
150 static int long_run_latency_avg = 0;
153 * Give user option to use default type for special cases -
154 * filestore.journal_wr_bytes is currently only metric here. Ceph reports the
155 * type as a sum/count pair and will calculate it the same as a latency value.
156 * All other "bytes" metrics (excluding the used/capacity bytes for the OSD)
157 * use the DERIVE type. Unless user specifies to use given type, convert this
158 * metric to use DERIVE.
160 static int convert_special_metrics = 1;
162 /** Array of daemons to monitor */
163 static struct ceph_daemon **g_daemons = NULL;
165 /** Number of elements in g_daemons */
166 static int g_num_daemons = 0;
169 * A set of data that we build up in memory while parsing the JSON.
173 /** ceph daemon we are processing data for*/
174 struct ceph_daemon *d;
175 /** track avgcount across counters for avgcount/sum latency pairs */
177 /** current index of counters - used to get type of counter */
179 /** do we already have an avgcount for latency pair */
182 * similar to index, but current index of latency type counters -
183 * used to get last poll data of counter
187 * values list - maintain across counters since
188 * host/plugin/plugin instance are always the same
194 * A set of count/sum pairs to keep track of latency types and get difference
195 * between this poll data and last poll data.
199 char ds_name[DATA_MAX_NAME_LEN];
204 /******* network I/O *******/
207 CSTATE_UNCONNECTED = 0,
208 CSTATE_WRITE_REQUEST,
216 ASOK_REQ_VERSION = 0,
219 ASOK_REQ_NONE = 1000,
224 /** The Ceph daemon that we're talking to */
225 struct ceph_daemon *d;
228 uint32_t request_type;
230 /** The connection state */
233 /** The socket we use to talk to this daemon */
236 /** The amount of data remaining to read / write. */
239 /** Length of the JSON to read */
242 /** Buffer containing JSON data */
245 /** Keep data important to yajl processing */
246 struct yajl_struct yajl;
249 static int ceph_cb_null(void *ctx)
251 return CEPH_CB_CONTINUE;
254 static int ceph_cb_boolean(void *ctx, int bool_val)
256 return CEPH_CB_CONTINUE;
259 #define BUFFER_ADD(dest, src) do { \
260 size_t dest_size = sizeof (dest); \
261 strncat ((dest), (src), dest_size - strlen (dest)); \
262 (dest)[dest_size - 1] = 0; \
266 ceph_cb_number(void *ctx, const char *number_val, yajl_len_t number_len)
268 yajl_struct *state = (yajl_struct*) ctx;
269 char buffer[number_len+1];
270 char key[2 * DATA_MAX_NAME_LEN] = { 0 };
271 _Bool latency_type = 0;
275 memcpy(buffer, number_val, number_len);
276 buffer[sizeof(buffer) - 1] = 0;
278 for (i = 0; i < state->depth; i++)
280 if (state->stack[i] == NULL)
283 if (strlen (key) != 0)
284 BUFFER_ADD (key, ".");
285 BUFFER_ADD (key, state->stack[i]);
288 /* Special case for latency metrics. */
289 if ((strcmp ("avgcount", state->key) == 0)
290 || (strcmp ("sum", state->key) == 0))
294 /* Super-special case for filestore.journal_wr_bytes.avgcount: For
295 * some reason, Ceph schema encodes this as a count/sum pair while all
296 * other "Bytes" data (excluding used/capacity bytes for OSD space) uses
297 * a single "Derive" type. To spare further confusion, keep this KPI as
298 * the same type of other "Bytes". Instead of keeping an "average" or
299 * "rate", use the "sum" in the pair and assign that to the derive
301 if (convert_special_metrics && (state->depth >= 2)
302 && (strcmp("filestore", state->stack[state->depth - 2]) == 0)
303 && (strcmp("journal_wr_bytes", state->stack[state->depth - 1]) == 0)
304 && (strcmp("avgcount", state->key) == 0))
306 DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes");
307 return CEPH_CB_CONTINUE;
310 else /* not a latency type */
312 BUFFER_ADD (key, ".");
313 BUFFER_ADD (key, state->key);
316 status = state->handler(state->handler_arg, buffer, key);
317 if((status == RETRY_AVGCOUNT) && latency_type)
319 /* Add previously skipped part of the key, either "avgcount" or "sum",
321 BUFFER_ADD (key, ".");
322 BUFFER_ADD (key, state->key);
324 status = state->handler(state->handler_arg, buffer, key);
329 ERROR("ceph plugin: JSON handler failed with status %d.", status);
330 return CEPH_CB_ABORT;
333 return CEPH_CB_CONTINUE;
336 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
337 yajl_len_t string_len)
339 return CEPH_CB_CONTINUE;
342 static int ceph_cb_start_map(void *ctx)
344 yajl_struct *state = (yajl_struct*) ctx;
346 /* Push key to the stack */
347 if (state->depth == YAJL_MAX_DEPTH)
348 return CEPH_CB_ABORT;
350 state->stack[state->depth] = state->key;
354 return CEPH_CB_CONTINUE;
357 static int ceph_cb_end_map(void *ctx)
359 yajl_struct *state = (yajl_struct*) ctx;
361 /* Pop key from the stack */
362 if (state->depth == 0)
363 return CEPH_CB_ABORT;
367 state->key = state->stack[state->depth];
368 state->stack[state->depth] = NULL;
370 return CEPH_CB_CONTINUE;
374 ceph_cb_map_key(void *ctx, const unsigned char *key, yajl_len_t string_len)
376 yajl_struct *state = (yajl_struct*) ctx;
377 size_t sz = ((size_t) string_len) + 1;
380 state->key = malloc (sz);
381 if (state->key == NULL)
383 ERROR ("ceph plugin: malloc failed.");
384 return CEPH_CB_ABORT;
387 memmove (state->key, key, sz - 1);
388 state->key[sz - 1] = 0;
390 return CEPH_CB_CONTINUE;
393 static int ceph_cb_start_array(void *ctx)
395 return CEPH_CB_CONTINUE;
398 static int ceph_cb_end_array(void *ctx)
400 return CEPH_CB_CONTINUE;
403 static yajl_callbacks callbacks = {
417 static void ceph_daemon_print(const struct ceph_daemon *d)
419 DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path);
422 static void ceph_daemons_print(void)
425 for(i = 0; i < g_num_daemons; ++i)
427 ceph_daemon_print(g_daemons[i]);
431 static void ceph_daemon_free(struct ceph_daemon *d)
434 for(; i < d->last_idx; i++)
436 sfree(d->last_poll_data[i]);
438 sfree(d->last_poll_data);
439 d->last_poll_data = NULL;
441 for(i = 0; i < d->ds_num; i++)
443 sfree(d->ds_names[i]);
450 /* compact_ds_name removed the special characters ":", "_", "-" and "+" from the
451 * intput string. Characters following these special characters are capitalized.
452 * Trailing "+" and "-" characters are replaces with the strings "Plus" and
454 static int compact_ds_name (char *buffer, size_t buffer_size, char const *src)
459 size_t ptr_size = buffer_size;
460 _Bool append_plus = 0;
461 _Bool append_minus = 0;
463 if ((buffer == NULL) || (buffer_size <= strlen ("Minus")) || (src == NULL))
466 src_copy = strdup (src);
467 src_len = strlen(src);
469 /* Remove trailing "+" and "-". */
470 if (src_copy[src_len - 1] == '+')
474 src_copy[src_len] = 0;
476 else if (src_copy[src_len - 1] == '-')
480 src_copy[src_len] = 0;
483 /* Split at special chars, capitalize first character, append to buffer. */
484 char *dummy = src_copy;
486 char *save_ptr = NULL;
487 while ((token = strtok_r (dummy, ":_-+", &save_ptr)) != NULL)
493 token[0] = toupper ((int) token[0]);
495 assert (ptr_size > 1);
497 len = strlen (token);
502 assert (len < ptr_size);
504 sstrncpy (ptr, token, len + 1);
513 /* Append "Plus" or "Minus" if "+" or "-" has been stripped above. */
514 if (append_plus || append_minus)
516 char const *append = "Plus";
520 size_t offset = buffer_size - (strlen (append) + 1);
521 if (offset > strlen (buffer))
522 offset = strlen (buffer);
524 sstrncpy (buffer + offset, append, buffer_size - offset);
531 static _Bool has_suffix (char const *str, char const *suffix)
533 size_t str_len = strlen (str);
534 size_t suffix_len = strlen (suffix);
537 if (suffix_len > str_len)
539 offset = str_len - suffix_len;
541 if (strcmp (str + offset, suffix) == 0)
547 /* count_parts returns the number of elements a "foo.bar.baz" style key has. */
548 static size_t count_parts (char const *key)
551 size_t parts_num = 0;
553 for (ptr = key; ptr != NULL; ptr = strchr (ptr + 1, '.'))
560 * Parse key to remove "type" if this is for schema and initiate compaction
562 static int parse_keys (char *buffer, size_t buffer_size, const char *key_str)
564 char tmp[2 * buffer_size];
566 if (buffer == NULL || buffer_size == 0 || key_str == NULL || strlen (key_str) == 0)
569 if ((count_parts (key_str) > 2) && has_suffix (key_str, ".type"))
571 /* strip ".type" suffix iff the key has more than two parts. */
572 size_t sz = strlen (key_str) - strlen (".type") + 1;
574 if (sz > sizeof (tmp))
576 sstrncpy (tmp, key_str, sz);
580 sstrncpy (tmp, key_str, sizeof (tmp));
583 return compact_ds_name (buffer, buffer_size, tmp);
587 * while parsing ceph admin socket schema, save counter name and type for later
590 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
594 char ds_name[DATA_MAX_NAME_LEN] = { 0 };
596 if(convert_special_metrics)
599 * Special case for filestore:JournalWrBytes. For some reason, Ceph
600 * schema encodes this as a count/sum pair while all other "Bytes" data
601 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
602 * type. To spare further confusion, keep this KPI as the same type of
603 * other "Bytes". Instead of keeping an "average" or "rate", use the
604 * "sum" in the pair and assign that to the derive value.
606 if((strcmp(name,"filestore.journal_wr_bytes.type") == 0))
612 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
618 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
624 d->ds_names[d->ds_num] = malloc(DATA_MAX_NAME_LEN);
625 if(!d->ds_names[d->ds_num])
630 type = (pc_type & PERFCOUNTER_DERIVE) ? DSET_RATE :
631 ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
632 d->ds_types[d->ds_num] = type;
634 if (parse_keys(ds_name, sizeof (ds_name), name))
639 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN -1);
640 d->ds_num = (d->ds_num + 1);
645 /******* ceph_config *******/
646 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
649 if(item->values_num != 1)
653 if(item->values[0].type != OCONFIG_TYPE_STRING)
657 val = item->values[0].value.string;
658 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
660 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
662 return -ENAMETOOLONG;
667 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
669 if(item->values_num != 1)
674 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
679 *dest = (item->values[0].value.boolean) ? 1 : 0;
683 static int cc_add_daemon_config(oconfig_item_t *ci)
686 struct ceph_daemon *nd, cd = { 0 };
687 struct ceph_daemon **tmp;
689 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
691 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
696 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
702 for(i=0; i < ci->children_num; i++)
704 oconfig_item_t *child = ci->children + i;
706 if(strcasecmp("SocketPath", child->key) == 0)
708 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
716 WARNING("ceph plugin: ignoring unknown option %s", child->key);
719 if(cd.name[0] == '\0')
721 ERROR("ceph plugin: you must configure a daemon name.\n");
724 else if(cd.asok_path[0] == '\0')
726 ERROR("ceph plugin(name=%s): you must configure an administrative "
727 "socket path.\n", cd.name);
730 else if(!((cd.asok_path[0] == '/') ||
731 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
733 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
734 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
738 tmp = realloc(g_daemons, (g_num_daemons+1) * sizeof(*g_daemons));
741 /* The positive return value here indicates that this is a
742 * runtime error, not a configuration error. */
747 nd = malloc(sizeof (*nd));
752 memcpy(nd, &cd, sizeof(*nd));
753 g_daemons[g_num_daemons++] = nd;
757 static int ceph_config(oconfig_item_t *ci)
761 for(i = 0; i < ci->children_num; ++i)
763 oconfig_item_t *child = ci->children + i;
764 if(strcasecmp("Daemon", child->key) == 0)
766 ret = cc_add_daemon_config(child);
769 ERROR("ceph plugin: Couldn't allocate memory");
774 //process other daemons and ignore this one
778 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
780 ret = cc_handle_bool(child, &long_run_latency_avg);
786 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
788 ret = cc_handle_bool(child, &convert_special_metrics);
796 WARNING("ceph plugin: ignoring unknown option %s", child->key);
803 * Parse JSON and get error message if present
806 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
808 yajl_status status = yajl_parse(hand, json, json_len);
813 case yajl_status_error:
814 msg = yajl_get_error(hand, /* verbose = */ 1,
815 /* jsonText = */ (unsigned char *) json,
816 (unsigned int) json_len);
817 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
818 yajl_free_error(hand, msg);
820 case yajl_status_client_canceled:
828 * Add entry for each counter while parsing schema
831 node_handler_define_schema(void *arg, const char *val, const char *key)
833 struct ceph_daemon *d = (struct ceph_daemon *) arg;
836 return ceph_daemon_add_ds_entry(d, key, pc_type);
840 * Latency counter does not yet have an entry in last poll data - add it.
842 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
845 d->last_poll_data[d->last_idx] = malloc(sizeof (*d->last_poll_data[d->last_idx]));
846 if(!d->last_poll_data[d->last_idx])
850 sstrncpy(d->last_poll_data[d->last_idx]->ds_name,ds_n,
851 sizeof(d->last_poll_data[d->last_idx]->ds_name));
852 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
853 d->last_poll_data[d->last_idx]->last_count = cur_count;
854 d->last_idx = (d->last_idx + 1);
859 * Update latency counter or add new entry if it doesn't exist
861 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
862 double cur_sum, uint64_t cur_count)
864 if((d->last_idx > index) && (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0))
866 d->last_poll_data[index]->last_sum = cur_sum;
867 d->last_poll_data[index]->last_count = cur_count;
871 if(!d->last_poll_data)
873 d->last_poll_data = malloc(sizeof (*d->last_poll_data));
874 if(!d->last_poll_data)
881 struct last_data **tmp_last = realloc(d->last_poll_data,
882 ((d->last_idx+1) * sizeof(struct last_data *)));
887 d->last_poll_data = tmp_last;
889 return add_last(d, ds_n, cur_sum, cur_count);
893 * If using index guess failed (shouldn't happen, but possible if counters
894 * get rearranged), resort to searching for counter name
896 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n)
899 for(; i < d->last_idx; i++)
901 if(strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0)
910 * Calculate average b/t current data and last poll data
911 * if last poll data exists
913 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
914 double cur_sum, uint64_t cur_count)
916 double result = -1.1, sum_delt = 0.0;
917 uint64_t count_delt = 0;
919 if(d->last_idx > index)
921 if(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)
925 //test previous index
926 else if((index > 0) && (strcmp(d->last_poll_data[index-1]->ds_name, ds_n) == 0))
928 tmp_index = (index - 1);
932 tmp_index = backup_search_for_last_avg(d, ds_n);
935 if((tmp_index > -1) && (cur_count > d->last_poll_data[tmp_index]->last_count))
937 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
938 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
939 result = (sum_delt / count_delt);
947 if(update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM)
955 * If using index guess failed, resort to searching for counter name
957 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
960 for(; idx < d->ds_num; idx++)
962 if(strcmp(d->ds_names[idx], ds_name) == 0)
964 return d->ds_types[idx];
967 return DSET_TYPE_UNFOUND;
971 * Process counter data and dispatch values
973 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
978 struct values_tmp *vtmp = (struct values_tmp*) arg;
979 uint32_t type = DSET_TYPE_UNFOUND;
980 int index = vtmp->index;
982 char ds_name[DATA_MAX_NAME_LEN] = { 0 };
984 if (parse_keys (ds_name, sizeof (ds_name), key))
989 if(index >= vtmp->d->ds_num)
991 //don't overflow bounds of array
992 index = (vtmp->d->ds_num - 1);
996 * counters should remain in same order we parsed schema... we maintain the
997 * index variable to keep track of current point in list of counters. first
998 * use index to guess point in array for retrieving type. if that doesn't
999 * work, use the old way to get the counter type
1001 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
1004 type = vtmp->d->ds_types[index];
1006 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
1009 type = vtmp->d->ds_types[index-1];
1012 if(type == DSET_TYPE_UNFOUND)
1014 //couldn't find right type by guessing, check the old way
1015 type = backup_search_for_type(vtmp->d, ds_name);
1021 if(vtmp->avgcount_exists == -1)
1023 sscanf(val, "%" PRIu64, &vtmp->avgcount);
1024 vtmp->avgcount_exists = 0;
1025 //return after saving avgcount - don't dispatch value
1026 //until latency calculation
1032 sscanf(val, "%lf", &sum);
1034 if(vtmp->avgcount == 0)
1039 /** User wants latency values as long run avg */
1040 if(long_run_latency_avg)
1042 result = (sum / vtmp->avgcount);
1046 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, vtmp->avgcount);
1047 if(result == -ENOMEM)
1054 vtmp->avgcount_exists = -1;
1055 vtmp->latency_index = (vtmp->latency_index + 1);
1059 sscanf(val, "%lf", &tmp_d);
1063 sscanf(val, "%" PRIu64, &tmp_u);
1066 case DSET_TYPE_UNFOUND:
1068 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1072 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1073 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1074 vtmp->vlist.values = &uv;
1075 vtmp->vlist.values_len = 1;
1077 vtmp->index = (vtmp->index + 1);
1078 plugin_dispatch_values(&vtmp->vlist);
1083 static int cconn_connect(struct cconn *io)
1085 struct sockaddr_un address = { 0 };
1087 if(io->state != CSTATE_UNCONNECTED)
1089 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1092 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1096 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1097 "failed: error %d", err);
1100 address.sun_family = AF_UNIX;
1101 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1104 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1107 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1113 flags = fcntl(fd, F_GETFL, 0);
1114 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1117 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1123 io->state = CSTATE_WRITE_REQUEST;
1130 static void cconn_close(struct cconn *io)
1132 io->state = CSTATE_UNCONNECTED;
1136 RETRY_ON_EINTR(res, close(io->asok));
1145 /* Process incoming JSON counter data */
1147 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1150 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1156 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1157 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1158 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1159 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1162 vtmp->avgcount_exists = -1;
1163 vtmp->latency_index = 0;
1165 yajl->handler_arg = vtmp;
1166 ret = traverse_json(io->json, io->json_len, hand);
1172 * Initiate JSON parsing and print error if one occurs
1174 static int cconn_process_json(struct cconn *io)
1176 if((io->request_type != ASOK_REQ_DATA) &&
1177 (io->request_type != ASOK_REQ_SCHEMA))
1186 hand = yajl_alloc(&callbacks,
1188 /* alloc funcs = */ NULL,
1190 /* alloc funcs = */ NULL, NULL,
1192 /* context = */ (void *)(&io->yajl));
1196 ERROR ("ceph plugin: yajl_alloc failed.");
1202 switch(io->request_type)
1205 io->yajl.handler = node_handler_fetch_data;
1206 result = cconn_process_data(io, &io->yajl, hand);
1208 case ASOK_REQ_SCHEMA:
1209 //init daemon specific variables
1211 io->d->last_idx = 0;
1212 io->d->last_poll_data = NULL;
1213 io->yajl.handler = node_handler_define_schema;
1214 io->yajl.handler_arg = io->d;
1215 result = traverse_json(io->json, io->json_len, hand);
1225 status = yajl_complete_parse(hand);
1227 status = yajl_parse_complete(hand);
1230 if (status != yajl_status_ok)
1232 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1233 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1234 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1236 yajl_free_error (hand, errmsg);
1246 static int cconn_validate_revents(struct cconn *io, int revents)
1248 if(revents & POLLERR)
1250 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1256 case CSTATE_WRITE_REQUEST:
1257 return (revents & POLLOUT) ? 0 : -EINVAL;
1258 case CSTATE_READ_VERSION:
1259 case CSTATE_READ_AMT:
1260 case CSTATE_READ_JSON:
1261 return (revents & POLLIN) ? 0 : -EINVAL;
1263 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1264 "illegal state on line %d", io->d->name, __LINE__);
1269 /** Handle a network event for a connection */
1270 static int cconn_handle_event(struct cconn *io)
1275 case CSTATE_UNCONNECTED:
1276 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1277 "state on line %d", io->d->name, __LINE__);
1280 case CSTATE_WRITE_REQUEST:
1283 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1284 io->request_type, "\" }\n");
1285 size_t cmd_len = strlen(cmd);
1287 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1288 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1289 io->d->name, io->state, io->amt, ret);
1295 if(io->amt >= cmd_len)
1298 switch (io->request_type)
1300 case ASOK_REQ_VERSION:
1301 io->state = CSTATE_READ_VERSION;
1304 io->state = CSTATE_READ_AMT;
1310 case CSTATE_READ_VERSION:
1313 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1314 sizeof(io->d->version) - io->amt));
1315 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1316 io->d->name, io->state, ret);
1322 if(io->amt >= sizeof(io->d->version))
1324 io->d->version = ntohl(io->d->version);
1325 if(io->d->version != 1)
1327 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1328 "expecting version %d!", io->d->name, io->d->version);
1331 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1332 "version %d", io->d->name, io->d->version);
1335 io->request_type = ASOK_REQ_SCHEMA;
1339 case CSTATE_READ_AMT:
1342 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1343 sizeof(io->json_len) - io->amt));
1344 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1345 io->d->name, io->state, ret);
1351 if(io->amt >= sizeof(io->json_len))
1353 io->json_len = ntohl(io->json_len);
1355 io->state = CSTATE_READ_JSON;
1356 io->json = calloc(1, io->json_len + 1);
1359 ERROR("ceph plugin: error callocing io->json");
1365 case CSTATE_READ_JSON:
1368 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1369 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1370 io->d->name, io->state, ret);
1376 if(io->amt >= io->json_len)
1378 ret = cconn_process_json(io);
1384 io->request_type = ASOK_REQ_NONE;
1389 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1390 "state on line %d", io->d->name, __LINE__);
1395 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1398 if(io->request_type == ASOK_REQ_NONE)
1400 /* The request has already been serviced. */
1403 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1405 /* If there are no counters to report on, don't bother
1412 case CSTATE_UNCONNECTED:
1413 ret = cconn_connect(io);
1423 fds->events = POLLOUT;
1425 case CSTATE_WRITE_REQUEST:
1427 fds->events = POLLOUT;
1429 case CSTATE_READ_VERSION:
1430 case CSTATE_READ_AMT:
1431 case CSTATE_READ_JSON:
1433 fds->events = POLLIN;
1436 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1437 "on line %d", io->d->name, __LINE__);
1442 /** Returns the difference between two struct timevals in milliseconds.
1443 * On overflow, we return max/min int.
1445 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1448 int sec_diff = t1->tv_sec - t2->tv_sec;
1449 int usec_diff = t1->tv_usec - t2->tv_usec;
1450 ret = usec_diff / 1000;
1451 ret += (sec_diff * 1000);
1452 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1455 /** This handles the actual network I/O to talk to the Ceph daemons.
1457 static int cconn_main_loop(uint32_t request_type)
1459 int i, ret, some_unreachable = 0;
1460 struct timeval end_tv;
1461 struct cconn io_array[g_num_daemons];
1463 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %d)", request_type);
1465 /* create cconn array */
1466 memset(io_array, 0, sizeof(io_array));
1467 for(i = 0; i < g_num_daemons; ++i)
1469 io_array[i].d = g_daemons[i];
1470 io_array[i].request_type = request_type;
1471 io_array[i].state = CSTATE_UNCONNECTED;
1474 /** Calculate the time at which we should give up */
1475 gettimeofday(&end_tv, NULL);
1476 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1482 struct cconn *polled_io_array[g_num_daemons];
1483 struct pollfd fds[g_num_daemons];
1484 memset(fds, 0, sizeof(fds));
1486 for(i = 0; i < g_num_daemons; ++i)
1488 struct cconn *io = io_array + i;
1489 ret = cconn_prepare(io, fds + nfds);
1492 WARNING("ceph plugin: cconn_prepare(name=%s,i=%d,st=%d)=%d",
1493 io->d->name, i, io->state, ret);
1495 io->request_type = ASOK_REQ_NONE;
1496 some_unreachable = 1;
1500 polled_io_array[nfds++] = io_array + i;
1509 gettimeofday(&tv, NULL);
1510 diff = milli_diff(&end_tv, &tv);
1515 WARNING("ceph plugin: cconn_main_loop: timed out.");
1518 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1521 ERROR("ceph plugin: poll(2) error: %d", ret);
1524 for(i = 0; i < nfds; ++i)
1526 struct cconn *io = polled_io_array[i];
1527 int revents = fds[i].revents;
1532 else if(cconn_validate_revents(io, revents))
1534 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1535 "revents validation error: "
1536 "revents=0x%08x", io->d->name, i, io->state, revents);
1538 io->request_type = ASOK_REQ_NONE;
1539 some_unreachable = 1;
1543 ret = cconn_handle_event(io);
1546 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1547 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1549 io->request_type = ASOK_REQ_NONE;
1550 some_unreachable = 1;
1555 done: for(i = 0; i < g_num_daemons; ++i)
1557 cconn_close(io_array + i);
1559 if(some_unreachable)
1561 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1565 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1570 static int ceph_read(void)
1572 return cconn_main_loop(ASOK_REQ_DATA);
1575 /******* lifecycle *******/
1576 static int ceph_init(void)
1579 ceph_daemons_print();
1581 ret = cconn_main_loop(ASOK_REQ_VERSION);
1583 return (ret) ? ret : 0;
1586 static int ceph_shutdown(void)
1589 for(i = 0; i < g_num_daemons; ++i)
1591 ceph_daemon_free(g_daemons[i]);
1596 DEBUG("ceph plugin: finished ceph_shutdown");
1600 void module_register(void)
1602 plugin_register_complex_config("ceph", ceph_config);
1603 plugin_register_init("ceph", ceph_init);
1604 plugin_register_read("ceph", ceph_read);
1605 plugin_register_shutdown("ceph", ceph_shutdown);
1607 /* vim: set sw=4 sts=4 et : */