• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* MIT License
2  *
3  * Copyright (c) 2024 Brad House
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a copy
6  * of this software and associated documentation files (the "Software"), to deal
7  * in the Software without restriction, including without limitation the rights
8  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9  * copies of the Software, and to permit persons to whom the Software is
10  * furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice (including the next
13  * paragraph) shall be included in all copies or substantial portions of the
14  * Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  *
24  * SPDX-License-Identifier: MIT
25  */
26 
27 
28 /* IMPLEMENTATION NOTES
29  * ====================
30  *
31  * With very little effort we should be able to determine fairly proper timeouts
32  * we can use based on prior query history.  We track in order to be able to
33  * auto-scale when network conditions change (e.g. maybe there is a provider
34  * failover and timings change due to that).  Apple appears to do this within
35  * their system resolver in MacOS.  Obviously we should have a minimum, maximum,
36  * and initial value to make sure the algorithm doesn't somehow go off the
37  * rails.
38  *
39  * Values:
40  * - Minimum Timeout: 250ms (approximate RTT half-way around the globe)
41  * - Maximum Timeout: 5000ms (Recommended timeout in RFC 1123), can be reduced
42  *   by ARES_OPT_MAXTIMEOUTMS, but otherwise the bound specified by the option
43  *   caps the retry timeout.
44  * - Initial Timeout: User-specified via configuration or ARES_OPT_TIMEOUTMS
45  * - Average latency multiplier: 5x (a local DNS server returning a cached value
46  *   will be quicker than if it needs to recurse so we need to account for this)
47  * - Minimum Count for Average: 3.  This is the minimum number of queries we
48  *   need to form an average for the bucket.
49  *
50  * Per-server buckets for tracking latency over time (these are ephemeral
51  * meaning they don't persist once a channel is destroyed).  We record both the
52  * current timespan for the bucket and the immediate preceding timespan in case
53  * of roll-overs we can still maintain recent metrics for calculations:
54  * - 1 minute
55  * - 15 minutes
56  * - 1 hr
57  * - 1 day
58  * - since inception
59  *
60  * Each bucket would contain:
61  * - timestamp (divided by interval)
62  * - minimum latency
63  * - maximum latency
64  * - total time
65  * - count
66  * NOTE: average latency is (total time / count), we will calculate this
67  *       dynamically when needed
68  *
69  * Basic algorithm for calculating timeout to use would be:
70  * - Scan from most recent bucket to least recent
71  * - Check timestamp of bucket, if doesn't match current time, continue to next
72  *   bucket
73  * - Check count of bucket, if its not at least the "Minimum Count for Average",
74  *   check the previous bucket, otherwise continue to next bucket
75  * - If we reached the end with no bucket match, use "Initial Timeout"
76  * - If bucket is selected, take ("total time" / count) as Average latency,
77  *   multiply by "Average Latency Multiplier", bound by "Minimum Timeout" and
78  *   "Maximum Timeout"
79  * NOTE: The timeout calculated may not be the timeout used.  If we are retrying
80  * the query on the same server another time, then it will use a larger value
81  *
82  * On each query reply where the response is legitimate (proper response or
83  * NXDOMAIN) and not something like a server error:
84  * - Cycle through each bucket in order
85  * - Check timestamp of bucket against current timestamp, if out of date
86  *   overwrite previous entry with values, clear current values
87  * - Compare current minimum and maximum recorded latency against query time and
88  *   adjust if necessary
89  * - Increment "count" by 1 and "total time" by the query time
90  *
91  * Other Notes:
92  * - This is always-on, the only user-configurable value is the initial
93  *   timeout which will simply re-uses the current option.
94  * - Minimum and Maximum latencies for a bucket are currently unused but are
95  *   there in case we find a need for them in the future.
96  */
97 
98 #include "ares_private.h"
99 
100 /*! Minimum timeout value. Chosen due to it being approximately RTT half-way
101  *  around the world */
102 #define MIN_TIMEOUT_MS 250
103 
104 /*! Multiplier to apply to average latency to come up with an initial timeout */
105 #define AVG_TIMEOUT_MULTIPLIER 5
106 
107 /*! Upper timeout bounds, only used if channel->maxtimeout not set */
108 #define MAX_TIMEOUT_MS 5000
109 
110 /*! Minimum queries required to form an average */
111 #define MIN_COUNT_FOR_AVERAGE 3
112 
ares_metric_timestamp(ares_server_bucket_t bucket,const ares_timeval_t * now,ares_bool_t is_previous)113 static time_t ares_metric_timestamp(ares_server_bucket_t  bucket,
114                                     const ares_timeval_t *now,
115                                     ares_bool_t           is_previous)
116 {
117   time_t divisor = 1; /* Silence bogus MSVC warning by setting default value */
118 
119   switch (bucket) {
120     case ARES_METRIC_1MINUTE:
121       divisor = 60;
122       break;
123     case ARES_METRIC_15MINUTES:
124       divisor = 15 * 60;
125       break;
126     case ARES_METRIC_1HOUR:
127       divisor = 60 * 60;
128       break;
129     case ARES_METRIC_1DAY:
130       divisor = 24 * 60 * 60;
131       break;
132     case ARES_METRIC_INCEPTION:
133       return is_previous ? 0 : 1;
134     case ARES_METRIC_COUNT:
135       return 0; /* Invalid! */
136   }
137 
138   if (is_previous) {
139     if (divisor >= now->sec) {
140       return 0;
141     }
142     return (time_t)((now->sec - divisor) / divisor);
143   }
144 
145   return (time_t)(now->sec / divisor);
146 }
147 
ares_metrics_record(const ares_query_t * query,ares_server_t * server,ares_status_t status,const ares_dns_record_t * dnsrec)148 void ares_metrics_record(const ares_query_t *query, ares_server_t *server,
149                          ares_status_t status, const ares_dns_record_t *dnsrec)
150 {
151   ares_timeval_t       now;
152   ares_timeval_t       tvdiff;
153   unsigned int         query_ms;
154   ares_dns_rcode_t     rcode;
155   ares_server_bucket_t i;
156 
157   if (status != ARES_SUCCESS) {
158     return;
159   }
160 
161   if (server == NULL) {
162     return;
163   }
164 
165   ares_tvnow(&now);
166 
167   rcode = ares_dns_record_get_rcode(dnsrec);
168   if (rcode != ARES_RCODE_NOERROR && rcode != ARES_RCODE_NXDOMAIN) {
169     return;
170   }
171 
172   ares_timeval_diff(&tvdiff, &query->ts, &now);
173   query_ms = (unsigned int)((tvdiff.sec * 1000) + (tvdiff.usec / 1000));
174   if (query_ms == 0) {
175     query_ms = 1;
176   }
177 
178   /* Place in each bucket */
179   for (i = 0; i < ARES_METRIC_COUNT; i++) {
180     time_t ts = ares_metric_timestamp(i, &now, ARES_FALSE);
181 
182     /* Copy metrics to prev and clear */
183     if (ts != server->metrics[i].ts) {
184       server->metrics[i].prev_ts          = server->metrics[i].ts;
185       server->metrics[i].prev_total_ms    = server->metrics[i].total_ms;
186       server->metrics[i].prev_total_count = server->metrics[i].total_count;
187       server->metrics[i].ts               = ts;
188       server->metrics[i].latency_min_ms   = 0;
189       server->metrics[i].latency_max_ms   = 0;
190       server->metrics[i].total_ms         = 0;
191       server->metrics[i].total_count      = 0;
192     }
193 
194     if (server->metrics[i].latency_min_ms == 0 ||
195         server->metrics[i].latency_min_ms > query_ms) {
196       server->metrics[i].latency_min_ms = query_ms;
197     }
198 
199     if (query_ms > server->metrics[i].latency_max_ms) {
200       server->metrics[i].latency_max_ms = query_ms;
201     }
202 
203     server->metrics[i].total_count++;
204     server->metrics[i].total_ms += (ares_uint64_t)query_ms;
205   }
206 }
207 
ares_metrics_server_timeout(const ares_server_t * server,const ares_timeval_t * now)208 size_t ares_metrics_server_timeout(const ares_server_t  *server,
209                                    const ares_timeval_t *now)
210 {
211   const ares_channel_t *channel = server->channel;
212   ares_server_bucket_t  i;
213   size_t                timeout_ms = 0;
214   size_t                max_timeout_ms;
215 
216   for (i = 0; i < ARES_METRIC_COUNT; i++) {
217     time_t ts = ares_metric_timestamp(i, now, ARES_FALSE);
218 
219     /* This ts has been invalidated, see if we should use the previous
220      * time period */
221     if (ts != server->metrics[i].ts ||
222         server->metrics[i].total_count < MIN_COUNT_FOR_AVERAGE) {
223       time_t prev_ts = ares_metric_timestamp(i, now, ARES_TRUE);
224       if (prev_ts != server->metrics[i].prev_ts ||
225           server->metrics[i].prev_total_count < MIN_COUNT_FOR_AVERAGE) {
226         /* Move onto next bucket */
227         continue;
228       }
229       /* Calculate average time for previous bucket */
230       timeout_ms = (size_t)(server->metrics[i].prev_total_ms /
231                             server->metrics[i].prev_total_count);
232     } else {
233       /* Calculate average time for current bucket*/
234       timeout_ms =
235         (size_t)(server->metrics[i].total_ms / server->metrics[i].total_count);
236     }
237 
238     /* Multiply average by constant to get timeout value */
239     timeout_ms *= AVG_TIMEOUT_MULTIPLIER;
240     break;
241   }
242 
243   /* If we're here, that means its the first query for the server, so we just
244    * use the initial default timeout */
245   if (timeout_ms == 0) {
246     timeout_ms = channel->timeout;
247   }
248 
249   /* don't go below lower bounds */
250   if (timeout_ms < MIN_TIMEOUT_MS) {
251     timeout_ms = MIN_TIMEOUT_MS;
252   }
253 
254   /* don't go above upper bounds */
255   max_timeout_ms = channel->maxtimeout ? channel->maxtimeout : MAX_TIMEOUT_MS;
256   if (timeout_ms > max_timeout_ms) {
257     timeout_ms = max_timeout_ms;
258   }
259 
260   return timeout_ms;
261 }
262