Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
Newer
Older
100644 471 lines (342 sloc) 18.697 kb
0c1b88b lots more docs.
Robey Pointer authored
1
62482e2 cleanup.
Robey Pointer authored
2 A working guide to kestrel
3 ==========================
0c1b88b lots more docs.
Robey Pointer authored
4
5 Kestrel is a very simple message queue that runs on the JVM and uses the
6 memcache protocol (with some extensions) to talk to clients.
7
8 A single kestrel server has a set of queues identified by a name, which is
9 also the filename of that queue's journal file (usually in
10 `/var/spool/kestrel`). Each queue is a strictly-ordered FIFO of "items" of
11 binary data. Usually this data is in some serialized format like JSON or
12 ruby's marshal format.
13
d6199a9 explain queue names.
Robey Pointer authored
14 Generally queue names should be limited to alphanumerics `[A-Za-z0-9]`, dash
15 (`-`) and underline (`_`). In practice, kestrel doesn't enforce any
16 restrictions other than the name can't contain slash (`/`) because that can't
17 be used in filenames, squiggle (`~`) because it's used for temporary files,
e45676a incorporate matt sanford's feedback.
Robey Pointer authored
18 plus (`+`) because it's used for fanout queues, and dot (`.`) because it's
19 reserved for future use. Queue names are case-sensitive, but if you're running
20 kestrel on OS X or Windows, you will want to refrain from taking advantage of
21 this, since the journal filenames on those two platforms are *not*
22 case-sensitive.
d6199a9 explain queue names.
Robey Pointer authored
23
0c1b88b lots more docs.
Robey Pointer authored
24 A cluster of kestrel servers is like a memcache cluster: the servers don't
25 know about each other, and don't do any cross-communication, so you can add as
26 many as you like. Clients have a list of all servers in the cluster, and pick
27 one at random for each operation. In this way, each queue appears to be spread
8c93d94 more doc improvements.
Robey Pointer authored
28 out across every server, with items in a loose ordering.
0c1b88b lots more docs.
Robey Pointer authored
29
58746d2 more docs!
Robey Pointer authored
30 When kestrel starts up, it scans the journal folder and creates queues based
31 on any journal files it finds there, to restore state to the way it was when
32 it last shutdown (or was killed or died). New queues are created by referring
33 to them (for example, adding or trying to remove an item). A queue can be
34 deleted with the "delete" command.
35
0c1b88b lots more docs.
Robey Pointer authored
36
37 Configuration
38 -------------
39
70ef6d0 add some more docs, and make the guide refer to them.
Robey Pointer authored
40 The config files for kestrel are scala expressions loaded at runtime, usually
41 from `production.scala`, although you can use `development.scala` by passing
42 `-Dstage=development` to the java command line.
43
44 The config file evaluates to a `KestrelConfig` object that's used to configure
45 the server as a whole, a default queue, and any overrides for specific named
46 queues. The fields on `KestrelConfig` are documented here with their default
47 values:
fd3f9d4 add more docs and clean them up a bit
Robey Pointer authored
48 [KestrelConfig.html](http://robey.github.com/kestrel/doc/main/api/net/lag/kestrel/config/KestrelConfig.html)
0c1b88b lots more docs.
Robey Pointer authored
49
50 To confirm the current configuration of each queue, send "dump_config" to
51 a server (which can be done over telnet).
52
53 To reload the config file on a running server, send "reload" the same way.
70ef6d0 add some more docs, and make the guide refer to them.
Robey Pointer authored
54 You should immediately see the changes in "dump_config", to confirm. Reloading
55 will only affect queue configuration, not global server configuration. To
56 change the server configuration, restart the server.
0c1b88b lots more docs.
Robey Pointer authored
57
70ef6d0 add some more docs, and make the guide refer to them.
Robey Pointer authored
58 Logging is configured according to `util-logging`. The logging configuration
59 syntax is described here:
fd3f9d4 add more docs and clean them up a bit
Robey Pointer authored
60 [util-logging](https://github.com/twitter/util/blob/master/util-logging/README.markdown)
0c1b88b lots more docs.
Robey Pointer authored
61
70ef6d0 add some more docs, and make the guide refer to them.
Robey Pointer authored
62 Per-queue configuration is documented here:
fd3f9d4 add more docs and clean them up a bit
Robey Pointer authored
63 [QueueBuilder.html](http://robey.github.com/kestrel/doc/main/api/net/lag/kestrel/config/QueueBuilder.html)
64
65
66 Full queues
67 -----------
68
69 A queue can have the following limits set on it:
70
71 - `maxItems` - total items in the queue
72 - `maxSize` - total bytes of data in the items in the queue
73
74 If either of these limits is reached, no new items can be added to the queue.
75 (Clients will receive an error when trying to add.) If you set
76 `discardOldWhenFull` to true, then all adds will succeed, and the oldest
77 item(s) will be silently discarded until the queue is back within the item
78 and size limits.
79
80 `maxItemSize` limits the size of any individual item. If an add is attempted
81 with an item larger than this limit, it always fails.
74f21d8 document expiration_timer_frequency_seconds, max_item_size, and move_…
Robey Pointer authored
82
0c1b88b lots more docs.
Robey Pointer authored
83
84 The journal file
85 ----------------
86
87 The journal file is the only on-disk storage of a queue's contents, and it's
88 just a sequential record of each add or remove operation that's happened on
89 that queue. When kestrel starts up, it replays each queue's journal to build
90 up the in-memory queue that it uses for client queries.
91
58746d2 more docs!
Robey Pointer authored
92 The journal file is rotated in one of two conditions:
0c1b88b lots more docs.
Robey Pointer authored
93
1531d0d finish cleaning up the guide.
Robey Pointer authored
94 1. the queue is empty and the journal is larger than `defaultJournalSize`
0c1b88b lots more docs.
Robey Pointer authored
95
1531d0d finish cleaning up the guide.
Robey Pointer authored
96 2. the journal is larger than `maxJournalSize`
0c1b88b lots more docs.
Robey Pointer authored
97
1531d0d finish cleaning up the guide.
Robey Pointer authored
98 For example, if `defaultJournalSize` is 16MB (the default), then if the queue
99 is empty and the journal is larger than 16MB, it will be truncated into a new
100 (empty) file. If the journal is larger than `maxJournalSize` (1GB by default),
101 the journal will be rewritten periodically to contain just the live items.
0c1b88b lots more docs.
Robey Pointer authored
102
1531d0d finish cleaning up the guide.
Robey Pointer authored
103 You can turn the journal off for a queue (`keepJournal` = false) and the queue
0c1b88b lots more docs.
Robey Pointer authored
104 will exist only in memory. If the server restarts, all enqueued items are
1531d0d finish cleaning up the guide.
Robey Pointer authored
105 lost. You can also force a queue's journal to be sync'd to disk periodically,
106 or even after every write operation, at a performance cost, using
107 `syncJournal`.
0c1b88b lots more docs.
Robey Pointer authored
108
1531d0d finish cleaning up the guide.
Robey Pointer authored
109 If a queue grows past `maxMemorySize` bytes (128MB by default), only the
0c1b88b lots more docs.
Robey Pointer authored
110 first 128MB is kept in memory. The journal is used to track later items, and
111 as items are removed, the journal is played forward to keep 128MB in memory.
112 This is usually known as "read-behind" mode, but Twitter engineers sometimes
113 refer to it as the "square snake" because of the diagram used to brainstorm
114 the implementation. When a queue is in read-behind mode, removing an item will
115 often cause 2 disk operations instead of one: one to record the remove, and
116 one to read an item in from disk to keep 128MB in memory. This is the
117 trade-off to avoid filling memory and crashing the JVM.
118
119
58746d2 more docs!
Robey Pointer authored
120 Item expiration
121 ---------------
0c1b88b lots more docs.
Robey Pointer authored
122
58746d2 more docs!
Robey Pointer authored
123 When they come from a client, expiration times are handled in the same way as
124 memcache: if the number is small (less than one million), it's interpreted as
125 a relative number of seconds from now. Otherwise it's interpreted as an
126 absolute unix epoch time, in seconds since the beginning of 1 January 1970
127 GMT.
0c1b88b lots more docs.
Robey Pointer authored
128
58746d2 more docs!
Robey Pointer authored
129 Expiration times are immediately translated into an absolute time, in
1531d0d finish cleaning up the guide.
Robey Pointer authored
130 *milliseconds*, and if it's further in the future than the queue's `maxAge`,
a3c112f fix typos
Robey Pointer authored
131 the `maxAge` is used instead. An expiration of 0, which is usually the
58746d2 more docs!
Robey Pointer authored
132 default, means an item never expires.
0c1b88b lots more docs.
Robey Pointer authored
133
58746d2 more docs!
Robey Pointer authored
134 Expired items are flushed from a queue whenever a new item is added or
fd3f9d4 add more docs and clean them up a bit
Robey Pointer authored
135 removed. Additionally, if the global config option `expirationTimerFrequency`
136 is set, a background thread will periodically remove expired items from the
137 head of each queue. The provided `production.conf` sets this to one second.
138 If this is turned off, an idle queue won't have any items expired, but you
139 can still trigger a check by doing a "peek" on it.
140
141 Normally, expired items are discarded. If `expireToQueue` is set, then
142 expired items are moved to the specified queue just as if a client had put
143 it there. The item is added with no expiration time, but that can be
144 overridden if the new queue has a default expiration policy.
145
146 To prevent stalling the server when it encounters a swarm of items that all
147 expired at the same time, `maxExpireSweep` limits the number of items that
148 will be removed by the background thread in a single round. This is primarily
149 useful as a throttling mechanism when using a queue as a way to delay work.
74f21d8 document expiration_timer_frequency_seconds, max_item_size, and move_…
Robey Pointer authored
150
7841145 @gphat Expiring queues.
gphat authored
151 Queue expiration
152 ----------------
153
154 Queues can be configure to expire whole queues as well. When a queue is
155 created the time is noted. It is periodically (see the aforementioned
156 `expirationTimerFrequency`) checked against the current time and the
157 `maxQueueAge` configuration option. If the queue is *empty* and the current time
158 is greater than create time + `maxQueueAge` then the queue is ready to be
159 expired and will be deleted.
160
161 A `maxQueueAge` of zero, which is usually the default, means a queue never
162 expires.
0c1b88b lots more docs.
Robey Pointer authored
163
58746d2 more docs!
Robey Pointer authored
164 Fanout Queues
165 -------------
166
167 If a queue name has a `+` in it (like "`orders+audit`"), it's treated as a
168 fanout queue, using the format `<parent>+<child>`. These queues belong to a
169 parent queue -- in this example, the "orders" queue. Every item written into
170 a parent queue will also be written into each of its children.
171
172 Fanout queues each have their own journal file (if the parent queue has a
173 journal file) and otherwise behave exactly like any other queue. You can get
174 and peek and even add items directly to a child queue if you want. It uses the
175 parent queue's configuration instead of having independent child queue
176 configuration blocks.
177
178 When a fanout queue is first referenced by a client, the journal file (if any)
179 is created, and it will start receiving new items written to the parent queue.
180 Existing items are not copied over. A fanout queue can be deleted to stop it
181 from receiving new items.
182
183
184 Memcache commands
185 -----------------
186
187 - `SET <queue-name> <flags (ignored)> <expiration> <# bytes>`
188
189 Add an item to a queue. It may fail if the queue has a size or item limit
190 and it's full.
191
e45676a incorporate matt sanford's feedback.
Robey Pointer authored
192 - `GET <queue-name>[options]`
66c498b docs!
Robey Pointer authored
193
58746d2 more docs!
Robey Pointer authored
194 Remove an item from a queue. It will return an empty response immediately if
195 the queue is empty. The queue name may be followed by options separated
196 by `/`:
197
66c498b docs!
Robey Pointer authored
198 - `/t=<milliseconds>`
199
200 Wait up to a given time limit for a new item to arrive. If an item arrives
201 on the queue within this timeout, it's returned as normal. Otherwise,
202 after that timeout, an empty response is returned.
203
204 - `/open`
205
206 Tentatively remove an item from the queue. The item is returned as usual
207 but is also set aside in case the client disappears before sending a
208 "close" request. (See "Reliable Reads" below.)
58746d2 more docs!
Robey Pointer authored
209
66c498b docs!
Robey Pointer authored
210 - `/close`
58746d2 more docs!
Robey Pointer authored
211
66c498b docs!
Robey Pointer authored
212 Close any existing open read. (See "Reliable Reads" below.)
58746d2 more docs!
Robey Pointer authored
213
66c498b docs!
Robey Pointer authored
214 - `/abort`
58746d2 more docs!
Robey Pointer authored
215
66c498b docs!
Robey Pointer authored
216 Cancel any existing open read, returing that item to the head of the
217 queue. It will be the next item fetched. (See "Reliable Reads" below.)
58746d2 more docs!
Robey Pointer authored
218
66c498b docs!
Robey Pointer authored
219 - `/peek`
58746d2 more docs!
Robey Pointer authored
220
66c498b docs!
Robey Pointer authored
221 Return the first available item from the queue, if there is one, but don't
222 remove it. You can't combine this with any of the reliable read options.
58746d2 more docs!
Robey Pointer authored
223
d1b4ee7 @rtyler Update guide.md
rtyler authored
224 For example, to open a new read, waiting up to 500msec for an item:
e45676a incorporate matt sanford's feedback.
Robey Pointer authored
225
226 GET work/t=500/open
227
228 Or to close an existing read and open a new one:
229
230 GET work/close/open
231
58746d2 more docs!
Robey Pointer authored
232 - `DELETE <queue-name>`
233
234 Drop a queue, discarding any items in it, and deleting any associated
235 journal files.
236
237 - `FLUSH <queue-name>`
238
239 Discard all items remaining in this queue. The queue remains live and new
240 items can be added. The time it takes to flush will be linear to the current
241 queue size, and any other activity on this queue will block while it's being
242 flushed.
243
244 - `FLUSH_ALL`
245
246 Discard all items remaining in all queues. The queues are flushed one at a
247 time, as if kestrel received a `FLUSH` command for each queue.
0c1b88b lots more docs.
Robey Pointer authored
248
58746d2 more docs!
Robey Pointer authored
249 - `VERSION`
250
251 Display the kestrel version in a way compatible with memcache.
252
253 - `SHUTDOWN`
254
255 Cleanly shutdown the server and exit.
256
257 - `RELOAD`
258
259 Reload the config file and reconfigure all queues. This should have no
260 noticable effect on the server's responsiveness.
261
262 - `DUMP_CONFIG`
263
264 Dump a list of each queue currently known to the server, and list the config
265 values for each queue. The format is:
266
66c498b docs!
Robey Pointer authored
267 queue 'master' {
268 max_items=2147483647
269 max_size=9223372036854775807
270 max_age=0
271 max_journal_size=16277216
272 max_memory_size=134217728
273 max_journal_overflow=10
274 max_journal_size_absolute=9223372036854775807
275 discard_old_when_full=false
276 journal=true
277 sync_journal=false
278 }
279
280 The last queue will be followed by `END` on a line by itself.
58746d2 more docs!
Robey Pointer authored
281
282 - `STATS`
283
284 Display server stats in memcache style. They're described below.
285
286 - `DUMP_STATS`
287
288 Display server stats in a more readable style, grouped by queue. They're
289 described below.
290
1531d0d finish cleaning up the guide.
Robey Pointer authored
291 - `MONITOR <queue-name> <seconds>`
292
293 Monitor a queue for a time, fetching any new items that arrive. Clients
294 are queued in a fair fashion, per-item, so many clients may monitor a
295 queue at once. After the given timeout, a separate `END` response will
296 signal the end of the monitor period. Any fetched items are open
297 transactions (see "Reliable Reads" below), and should be closed with
298 `CONFIRM`.
299
300 - `CONFIRM <queue-name> <count>`
301
302 Confirm receipt of `count` items from a queue. Usually this is the response
303 to a `MONITOR` command, to confirm the items that arrived during the monitor
304 period.
305
74f21d8 document expiration_timer_frequency_seconds, max_item_size, and move_…
Robey Pointer authored
306
58746d2 more docs!
Robey Pointer authored
307 Reliable reads
308 --------------
309
310 Normally when a client removes an item from the queue, kestrel immediately
311 discards the item and assumes the client has taken ownership. This isn't
312 always safe, because a client could crash or lose the network connection
313 before it gets the item. So kestrel also supports a "reliable read" that
314 happens in two stages, using the `/open` and `/close` options to `GET`.
315
316 When `/open` is used, and an item is available, kestrel will remove it from
317 the queue and send it to the client as usual. But it will also set the item
318 aside. If a client disconnects while it has an open read, the item is put back
e45676a incorporate matt sanford's feedback.
Robey Pointer authored
319 into the queue, at the head, so it will be the next item fetched. Only one
320 item can be "open" per client connection.
58746d2 more docs!
Robey Pointer authored
321
322 A previous open request is closed with `/close`. The server will reject any
323 attempt to open another read when one is already open, but it will ignore
324 `/close` if there's no open request, so that you can add `/close` to every
325 `GET` request for convenience.
326
327 If for some reason you want to abort a read without disconnecting, you can use
328 `/abort`. But because aborted items are placed back at the head of the queue,
329 this isn't a good way to deal with client errors. Since the error-causing item
330 will always be the next one available, you'll end up bouncing the same item
331 around between clients instead of making progress.
332
333 There's always a trade-off: either potentially lose items or potentially
334 receive the same item multiple times. Reliable reads choose the latter option.
335 To use this tactic successfully, work items should be idempotent, meaning the
336 work could be done 2 or 3 times and have the same effect as if it had been
337 done only once (except wasting some resources).
338
e45676a incorporate matt sanford's feedback.
Robey Pointer authored
339 Example:
340
341 GET dirty_jobs/close/open
342 (receives job 1)
343 GET dirty_jobs/close/open
344 (closes job 1, receives job 2)
345 ...etc...
58746d2 more docs!
Robey Pointer authored
346
74f21d8 document expiration_timer_frequency_seconds, max_item_size, and move_…
Robey Pointer authored
347
58746d2 more docs!
Robey Pointer authored
348 Server stats
349 ------------
350
66c498b docs!
Robey Pointer authored
351 Global stats reported by kestrel are:
352
353 - `uptime` - seconds the server has been online
354 - `time` - current time in unix epoch
355 - `version` - version string, like "1.2"
356 - `curr_items` - total of items waiting in all queues
357 - `total_itmes` - total of items that have ever been added in this server's
358 lifetime
359 - `bytes` - total byte size of items waiting in all queues
360 - `curr_connections` - current open connections from clients
361 - `total_connections` - total connections that have been opened in this
362 server's lifetime
363 - `cmd_get` - total `GET` requests
364 - `cmd_set` - total `SET` requests
365 - `cmd_peek` - total `GET/peek` requests
366 - `get_hits` - total `GET` requests that received an item
367 - `get_misses` - total `GET` requests on an empty queue
368 - `bytes_read` - total bytes read from clients
369 - `bytes_written` - total bytes written to clients
0d8a507 @gphat Add more statistics along with tests and docs.
gphat authored
370 - `queue_creates` - total number of queues created
7841145 @gphat Expiring queues.
gphat authored
371 - `queue_deletes` - total number of queues deleted (includes expires)
372 - `queue_expires` - total number of queues expires
66c498b docs!
Robey Pointer authored
373
374 For each queue, the following stats are also reported:
375
376 - `items` - items waiting in this queue
377 - `bytes` - total byte size of items waiting in this queue
378 - `total_items` - total items that have been added to this queue in this
379 server's lifetime
380 - `logsize` - byte size of the queue's journal file
381 - `expired_items` - total items that have been expired from this queue in this
382 server's lifetime
383 - `mem_items` - items in this queue that are currently in memory
384 - `mem_bytes` - total byte size of items in this queue that are currently in
385 memory (will always be less than or equal to `max_memory_size` config for
386 the queue)
387 - `age` - time, in milliseconds, that the last item to be fetched from this
388 queue had been waiting; that is, the time between `SET` and `GET`; if the
389 queue is empty, this will always be zero
390 - `discarded` - number of items discarded because the queue was too full
391 - `waiters` - number of clients waiting for an item from this queue (using
392 `GET/t`)
721bf6b document the new stat.
Robey Pointer authored
393 - `open_transactions` - items read with `/open` but not yet confirmed
0d8a507 @gphat Add more statistics along with tests and docs.
gphat authored
394 - `total_flushes` total number of times this queue has been flushed
66c498b docs!
Robey Pointer authored
395
396
397 Kestrel as a library
398 --------------------
399
400 You can use kestrel as a library by just sticking the jar on your classpath.
401 It's a cheap way to get a durable work queue for inter-process or inter-thread
402 communication. Each queue is represented by a `PersistentQueue` object:
403
1531d0d finish cleaning up the guide.
Robey Pointer authored
404 class PersistentQueue(val name: String, persistencePath: String,
405 @volatile var config: QueueConfig, timer: Timer,
406 queueLookup: Option[(String => Option[PersistentQueue])]) {
66c498b docs!
Robey Pointer authored
407
408 and must be initialized before using:
409
410 def setup(): Unit
58746d2 more docs!
Robey Pointer authored
411
66c498b docs!
Robey Pointer authored
412 specifying the path for the journal files (if the queue will be journaled),
1531d0d finish cleaning up the guide.
Robey Pointer authored
413 the name of the queue, a `QueueConfig` object (derived from `QueueBuilder`),
414 a timer for handling timeout reads, and optionally a way to find other named
415 queues (for `expireToQueue` support).
58746d2 more docs!
Robey Pointer authored
416
66c498b docs!
Robey Pointer authored
417 To add an item to a queue:
58746d2 more docs!
Robey Pointer authored
418
1531d0d finish cleaning up the guide.
Robey Pointer authored
419 def add(value: Array[Byte], expiry: Option[Time]): Boolean
58746d2 more docs!
Robey Pointer authored
420
66c498b docs!
Robey Pointer authored
421 It will return `false` if the item was rejected because the queue was full.
0c1b88b lots more docs.
Robey Pointer authored
422
66c498b docs!
Robey Pointer authored
423 Queue items are represented by a case class:
424
1531d0d finish cleaning up the guide.
Robey Pointer authored
425 case class QItem(addTime: Time, expiry: Option[Time], data: Array[Byte], var xid: Int)
66c498b docs!
Robey Pointer authored
426
427 and several operations exist to remove or peek at the head item:
428
429 def peek(): Option[QItem]
430 def remove(): Option[QItem]
431
432 To open a reliable read, set `transaction` true, and later confirm or unremove
433 the item by its `xid`:
434
435 def remove(transaction: Boolean): Option[QItem]
436 def unremove(xid: Int)
437 def confirmRemove(xid: Int)
438
1531d0d finish cleaning up the guide.
Robey Pointer authored
439 You can also asynchronously remove or peek at items using futures.
66c498b docs!
Robey Pointer authored
440
1531d0d finish cleaning up the guide.
Robey Pointer authored
441 def waitRemove(deadline: Option[Time], transaction: Boolean): Future[Option[QItem]]
442 def waitPeek(deadline: Option[Time]): Future[Option[QItem]]
66c498b docs!
Robey Pointer authored
443
444 When done, you should close the queue:
445
446 def close(): Unit
447 def isClosed: Boolean
448
449 Here's a short example:
450
1531d0d finish cleaning up the guide.
Robey Pointer authored
451 var queue = new PersistentQueue("work", "/var/spool/kestrel", config, timer, None)
66c498b docs!
Robey Pointer authored
452 queue.setup()
453
454 // add an item with no expiration:
455 queue.add("hello".getBytes, 0)
456
457 // start to remove it, then back out:
458 val item = queue.remove(true)
459 queue.unremove(item.xid)
460
461 // remove an item with a 500msec timeout, and confirm it:
1531d0d finish cleaning up the guide.
Robey Pointer authored
462 queue.waitRemove(500.milliseconds.fromNow, true)() match {
463 case None =>
464 println("nothing. :(")
465 case Some(item) =>
466 println("got: " + new String(item.data))
467 queue.confirmRemove(item.xid)
66c498b docs!
Robey Pointer authored
468 }
0c1b88b lots more docs.
Robey Pointer authored
469
66c498b docs!
Robey Pointer authored
470 queue.close()
Something went wrong with that request. Please try again.