OLD | NEW |
---|---|
(Empty) | |
1 // Copyright 2015 The Chromium Authors. All rights reserved. | |
2 // Use of this source code is governed by a BSD-style license that can be | |
3 // found in the LICENSE file. | |
4 | |
5 // Package dscache provides a transparent cache for RawDatastore which is | |
6 // backed by Memcache. | |
7 // | |
8 // Inspiration | |
9 // | |
10 // Although this is not a port of any particular implementation, it takes | |
11 // inspiration from these fine libraries: | |
12 // - https://cloud.google.com/appengine/docs/python/ndb/ | |
13 // - https://github.com/qedus/nds | |
14 // - https://github.com/mjibson/goon | |
15 // | |
16 // Algorithm | |
17 // | |
18 // Memcache contains cache entries for single datastore entities. The memcache | |
19 // key looks like | |
20 // | |
21 // "gae:" | vers | ":" | shard | ":" | Base64_std_nopad(SHA1(datastore.Key)) | |
22 // | |
23 // Where: | |
24 // - vers is an ascii-hex-encoded number (currently 1). | |
25 // - shard is a zero-based ascii-hex-encoded number (depends on shardsForKey). | |
26 // - SHA1 has been chosen as unlikely (p == 1e-18) to collide, given dedicated | |
27 // memcache sizes of up to 170 Exabytes (assuming an average entry size of | |
28 // 100KB including the memcache key). This is clearly overkill, but MD5 | |
29 // could start showing collisions at this probability in as small as a 26GB | |
30 // cache (and also MD5 sucks). | |
31 // | |
32 // The memcache value is a compression byte, indicating the scheme (See | |
33 // CompressionType), followed by the encoded (and possibly compressed) value. | |
34 // Encoding is done with datastore.PropertyMap.Write(). The memcache value | |
35 // may also be the empty byte sequence, indicating that this entity is deleted. | |
36 // | |
37 // The memcache entry may also have a 'flags' value set to one of the following: | |
38 // - 0 "entity" (cached value) | |
39 // - 1 "lock" (someone is mutating this entry) | |
40 // | |
41 // Algorithm - Put and Delete | |
42 // | |
43 // On a Put (or Delete), an empty value is unconditionally written to | |
44 // memcache with a LockTimeSeconds expiration (default 31 seconds), and | |
45 // a memcache flag value of 0x1 (indicating that it's a put-locked key). The | |
46 // random value is to preclude Get operations from believing that they possess | |
47 // the lock. | |
48 // | |
49 // NOTE: If this memcache Set fails, it's a HARD ERROR. See DANGER ZONE. | |
50 // | |
51 // The datastore operation will then occur. Assuming success, Put will then | |
52 // unconditionally delete all of the memcache locks. At some point later, a | |
53 // Get will write its own lock, get the value from datastore, and compare and | |
54 // swap to populate the value (detailed below). | |
55 // | |
56 // Algorithm - Get | |
57 // | |
58 // On a Get, "Add" a lock for it (which only does something if there's no entry | |
59 // in memcache yet) with a nonce value. We immediately Get the memcache entries | |
60 // back (for CAS purposes later). | |
61 // | |
62 // If it doesn't exist (unlikely since we just Add'd it) or if its flag is | |
63 // "lock" and the Value != the nonce we put there, go hit the datastore without | |
64 // trying to update memcache. | |
65 // | |
66 // If its flag is "entity", decode the object and return it. If the Value is | |
67 // the empty byte sequence, return ErrNoSuchEntity. | |
68 // | |
69 // If its flag is "lock" and the Value equals the nonce, go get it from the | |
70 // datastore. If that's successful, then encode the value to bytes, and CAS | |
71 // the object to memcache. The CAS will succeed if nothing else touched the | |
72 // memcache in the meantime (like a Put, a memcache expiration/eviction, etc.). | |
73 // | |
74 // Algorithm - Transactions | |
75 // | |
76 // In a transaction, all Put memcache operations are held until the very end of | |
77 // the transaction. Right before the transaction is committed, all accumulated | |
78 // Put memcache items are unconditionally set into memcache. | |
79 // | |
80 // NOTE: If this memcache Set fails, it's a HARD ERROR. See DANGER ZONE. | |
81 // | |
82 // If the transaction is sucessfully committed (err == nil), then all the locks | |
83 // will be deleted. | |
84 // | |
85 // The assumption here is that get operations apply all outstanding | |
86 // transactions before they return data (https://cloud.google.com/appengine/docs /go/datastore/#Go_Datastore_writes_and_data_visibility), | |
87 // and so it is safe to purge all the locks if the transaction is known-good. | |
88 // | |
89 // If the transaction succeeds, but RunInTransaction returns an error (which can | |
90 // happen), or if the transaction fails, then the lock entries time out | |
91 // naturally. This will mean 31-ish seconds of direct datastore access, but it's | |
92 // the more-correct thing to do. | |
93 // | |
94 // Gets and Queries in a transaction pass right through without reading or | |
95 // writing memcache. | |
96 // | |
97 // Cache control | |
98 // | |
99 // An entity may expose the following metadata (see | |
100 // datastore.PropertyLoadSaver.GetMeta) to control the behavior of its cache. | |
101 // | |
102 // - `gae:"$dscache.enable,<true|false>"` - whether or not this entity should | |
103 // be cached at all. If ommitted, dscache defaults to true. | |
104 // - `gae:"$dscache.expiration,#seconds"` - the number of seconds of | |
105 // persistance to use when this item is cached. 0 is infinite. If omitted, | |
106 // defaults to 0. | |
107 // | |
108 // In addition, the application may set a function shardsForKey(key) which | |
109 // returns the number of shards to use for a given datastore key. This function | |
110 // is set with the invocation of FilterRDS. | |
111 // | |
112 // Shards have the effect that all write (Put/Delete) operations clear all | |
113 // memcache entries for the given datastore entry, and all reads read (and | |
114 // possibly populate) one of the memcache entries. So if an entity has 4 shards, | |
115 // a datastore Get for it will pull from one of the 4 possible memcache keys | |
116 // at random. This is good for heavily-read, but infrequently updated, entities. | |
Vadim Sh.
2015/08/07 17:30:47
nit: mention that it's needed to avoid having hot
iannucci
2015/08/07 19:41:00
Done.
| |
117 // | |
118 // Caveats | |
119 // | |
120 // A couple things to note that may differ from other appengine datastore | |
121 // caching libraries (like goon, nds, or ndb). | |
122 // | |
123 // - It does NOT provide in-memory ("per-request") caching. | |
124 // - It's INtolerant of some memcache failures, but in exchange will not retur n | |
125 // inconsistent results. See DANGER ZONE for details. | |
126 // - Queries do not interact with the cache at all. | |
127 // - Negative lookups (e.g. ErrNoSuchEntity) are cached. | |
128 // - Allows sharding hot memcache entries as recommended by | |
129 // https://cloud.google.com/appengine/articles/best-practices-for-app-engine -memcache#distribute-load . | |
130 // | |
131 // DANGER ZONE | |
132 // | |
133 // As mentioned in the Put/Delete/Transactions sections above, if the memcache | |
134 // Set fails, that's a HARD ERROR. The reason for this is that otherwise in the | |
135 // event of transient memcache failures, the memcache may be permanently left in | |
136 // an inconsistent state, since there will be nothing to actually ensure that | |
137 // the bad value is flushed from memcache. As long as the Put is allowed to | |
138 // write the lock, then all will be (eventually) well, and so all other memcache | |
139 // operations are best effort. | |
140 // | |
141 // So, if memcache is DOWN, you will effectively see tons of errors in the logs, | |
142 // and all cached datastore access will be essentially degraded to a slow | |
143 // read-only state. At this point, you have essentially 3 mitigration | |
144 // strategies: | |
145 // - wait for memcache to come back up. | |
146 // - dynamically disable all memcache access by writing the datastore entry: | |
147 // /dscache,1 = {"Enable": false} | |
148 // in the default namespace. This can be done by invoking the | |
149 // SetDynamicGlobalEnable method. This can take up to 5 minutes to take | |
150 // effect. If you have very long-running backend requests, you may need to | |
151 // cycle them to have it take effect. This dynamic bit is read essentially | |
152 // once per http request (when FilteRDS is called on the context). | |
153 // - push a new version of the application disabling the cache filter | |
154 // by setting InstanceEnabledStatic to false in an init() function. | |
155 // | |
156 // On every dscache.FilterRDS invocation, it takes the opportunity to fetch this | |
157 // datastore value, if it hasn't been fetched in the last | |
158 // GlobalEnabledCheckInterval time (5 minutes). This equates to essentially once | |
159 // per http request, per 5 minutes, per instance. | |
160 package dscache | |
OLD | NEW |