|
| 1 | += Redis Objects - Lightweight object layer around redis-rb you can also use in a model |
| 2 | + |
| 3 | +Redis is great _not_ as a replacement for MySQL, but as a way to perform atomic |
| 4 | +operations on _individual_ data structures, like counters, lists, and sets. People |
| 5 | +that are wrapping ORM's around Redis are missing the point. |
| 6 | + |
| 7 | +This gem, instead, provides atomic methods that you can use *with* your existing |
| 8 | +ActiveRecord/DataMapper/etc models, or in classes that have nothing to do with an |
| 9 | +ORM or even a database, but need support for high-performance atomic operations. |
| 10 | + |
| 11 | +The only requirement Redis::Atoms has is that your class must provide an +id+ instance |
| 12 | +method which returns the ID for that instance. ActiveRecord, DataMapper, and MongoRecord |
| 13 | +all have id methods which are known to be suitable. Since +id+ can be anything as |
| 14 | +far as Redis::Atoms is concerned, you can even write an +id+ method of your own that |
| 15 | +just returns a string, or an MD5 of the name, or something else unique. |
| 16 | + |
| 17 | +== Installation |
| 18 | + |
| 19 | + gem install gemcutter |
| 20 | + gem tumble |
| 21 | + gem install redis-atoms |
| 22 | + |
| 23 | +== Example |
| 24 | + |
| 25 | +Somewhere in your app initialization |
| 26 | + |
| 27 | + require 'redis' |
| 28 | + require 'redis/atoms' |
| 29 | + Redis::Atoms.redis = Redis.new(:host => 127.0.0.1, :port => 6379) |
| 30 | + |
| 31 | +Model class: |
| 32 | + |
| 33 | + class Team < ActiveRecord::Base |
| 34 | + include Redis::Atoms |
| 35 | + |
| 36 | + counter :drafted_players |
| 37 | + counter :active_players |
| 38 | + lock :reorder, :timeout => 5 # seconds |
| 39 | + end |
| 40 | + |
| 41 | +Using counters to handle concurrency: |
| 42 | + |
| 43 | + @team = Team.find(1) |
| 44 | + if @team.drafted_players.increment <= @team.max_players |
| 45 | + # do stuff |
| 46 | + @team.team_players.create!(:player_id => 221) |
| 47 | + @team.active_players.increment |
| 48 | + else |
| 49 | + # reset counter state |
| 50 | + @team.drafted_players.decrement |
| 51 | + end |
| 52 | + |
| 53 | +Atomic block - a cleaner way to do the above. Exceptions or nil results |
| 54 | +rewind counter back to previous state: |
| 55 | + |
| 56 | + @team.drafted_players.increment do |val| |
| 57 | + raise Team::TeamFullError if val > @team.max_players |
| 58 | + @team.team_players.create!(:player_id => 221) |
| 59 | + @team.active_players.increment |
| 60 | + end |
| 61 | + |
| 62 | +Similar approach, using an if block (failure rewinds counter): |
| 63 | + |
| 64 | + @team.drafted_players.increment do |val| |
| 65 | + if val <= @team.max_players |
| 66 | + @team.team_players.create!(:player_id => 221) |
| 67 | + @team.active_players.increment |
| 68 | + end |
| 69 | + end |
| 70 | + |
| 71 | +Class methods work too - notice we override ActiveRecord counters: |
| 72 | + |
| 73 | + Team.increment_counter :drafted_players, team_id |
| 74 | + Team.decrement_counter :drafted_players, team_id, 2 |
| 75 | + |
| 76 | +Class-level atomic block (may save a DB fetch depending on your app): |
| 77 | + |
| 78 | + Team.increment_counter(:drafted_players, team_id) do |val| |
| 79 | + TeamPitcher.create!(:team_id => team_id, :pitcher_id => 181) |
| 80 | + Team.increment_counter(:active_players, team_id) |
| 81 | + end |
| 82 | + |
| 83 | +Locks with Redis. On completion or exception the lock is released: |
| 84 | + |
| 85 | + @team.reorder_lock.lock do |
| 86 | + @team.reorder_all_players |
| 87 | + end |
| 88 | + |
| 89 | +Class-level lock (same concept) |
| 90 | + |
| 91 | + Team.obtain_lock(:reorder, team_id) do |
| 92 | + Team.reorder_all_players(team_id) |
| 93 | + end |
| 94 | + |
| 95 | +== You Likely Have Some Huge Bugs |
| 96 | + |
| 97 | +You are probably not handling atomic operations properly in your app, and |
| 98 | +probably have some nasty lurking race conditions. The worst part is these |
| 99 | +will get worse as your user count increases, are difficult to reproduce, |
| 100 | +and usually happen to your most critical pieces of code. (And no, your |
| 101 | +rspec tests can't catch them either.) |
| 102 | + |
| 103 | +Let's assume you're writing an app to enable students to enroll in courses. |
| 104 | +You need to ensure that no more than 30 students can sign up for a given course. |
| 105 | +In your enrollment code, you have something like this: |
| 106 | + |
| 107 | + @course = Course.find(1) |
| 108 | + if @course.num_students < 30 |
| 109 | + @course.course_students.create!(:student_id => 101) |
| 110 | + @course.num_students += 1 |
| 111 | + @course.save! |
| 112 | + else |
| 113 | + # course is full |
| 114 | + end |
| 115 | + |
| 116 | +You're screwed. You now have 32 people in your 30 person class, and you have |
| 117 | +no idea what happened. |
| 118 | + |
| 119 | +"Well no duh," you're saying, "even the {ActiveRecord docs mention locking}[http://api.rubyonrails.org/classes/ActiveRecord/Locking/Pessimistic.html], |
| 120 | +so I'll just use that." |
| 121 | + |
| 122 | + @course = Course.find(1, :lock => true) |
| 123 | + if @course.num_students < 30 |
| 124 | + # ... |
| 125 | + |
| 126 | +Nice try, but now you've introduced other issues. Any other piece of code |
| 127 | +in your entire app that needs to update _anything_ about the course - maybe |
| 128 | +the course name, or start date, or location - is now serialized. If you need high |
| 129 | +concurrency, you're still screwed. |
| 130 | + |
| 131 | +You think, "ah-ha, the problem is having a separate counter!" |
| 132 | + |
| 133 | + @course = Course.find(1) |
| 134 | + if @course.course_students.count < 30 |
| 135 | + @course.course_students.create!(:student_id => 101) |
| 136 | + else |
| 137 | + # course is full |
| 138 | + end |
| 139 | + |
| 140 | +Nope. Still screwed. |
| 141 | + |
| 142 | +== The Root Down |
| 143 | + |
| 144 | +It's worth understanding the root issue, and how to address it. |
| 145 | + |
| 146 | +Race conditions arise from the difference in time between *evaluating* and *altering* |
| 147 | +a value. In our example, we fetched the record, then checked the value, then |
| 148 | +changed it. The more lines of code between those operations, and the higher your user |
| 149 | +count, the bigger the window of opportunity for other clients to get the data in an |
| 150 | +inconsistent state. |
| 151 | + |
| 152 | +Sometimes race conditions don't matter in practice, since often a user is |
| 153 | +only operating their own data. This has a race condition, but is probably ok: |
| 154 | + |
| 155 | + @post = Post.create(:user_id => @user.id, :title => "Whattup", ...) |
| 156 | + @user.total_posts += 1 # update my post count |
| 157 | + |
| 158 | +But this _would_ be problematic: |
| 159 | + |
| 160 | + @post = Post.create(:user_id => @user.id, :title => "Whattup", ...) |
| 161 | + @blog.total_posts += 1 # update post count across all users |
| 162 | + |
| 163 | +As multiple users could be adding posts concurrently. |
| 164 | + |
| 165 | +In a traditional RDBMS, you can increment counters atomically (but not return them) |
| 166 | +by firing off an update statement that self-references the column: |
| 167 | + |
| 168 | + update users set total_posts = total_posts + 1 where id = 372 |
| 169 | + |
| 170 | +You may have seen {ActiveRecord's increment_counter class method}[http://api.rubyonrails.org/classes/ActiveRecord/Base.html#M002278], |
| 171 | +which wraps this functionality. But outside of being cumbersome, this has |
| 172 | +the side effect that your object is no longer in sync with the DB, so you |
| 173 | +get other issues: |
| 174 | + |
| 175 | + Blog.increment_counter :total_posts, @blog.id |
| 176 | + if @blog.total_posts == 1000 |
| 177 | + # the 1000th poster - award them a gold star! |
| 178 | + |
| 179 | +The DB says 1000, but your @blog object still says 999, and the right person |
| 180 | +doesn't get their gold star. Sad faces all around. |
| 181 | + |
| 182 | +== A Better Way |
| 183 | + |
| 184 | +Bottom line: Any operation that could alter a value *must* return that value in |
| 185 | +the _same_ _operation_ for it to be atomic. If you do a separate get then set, |
| 186 | +or set then get, you're open to a race condition. There are very few systems that |
| 187 | +support an "increment and return" type operation, and Redis is one of them |
| 188 | +(Oracle sequences are another). |
| 189 | + |
| 190 | +When you think of the specific things that you need to ensure, many of these will |
| 191 | +reduce to numeric operations: |
| 192 | + |
| 193 | +* Ensuring there are no more than 30 students in a course |
| 194 | +* Getting more than 2 but less than 6 people in a game |
| 195 | +* Keeping a chat room to a max of 50 people |
| 196 | +* Correctly recording the total number of blog posts |
| 197 | +* Only allowing one piece of code to reorder a large dataset at a time |
| 198 | + |
| 199 | +All except the last one can be implemented with counters. The last one |
| 200 | +will need a carefully placed lock. |
| 201 | + |
| 202 | +The best way I've found to balance atomicity and concurrency is, for each value, |
| 203 | +actually create two counters: |
| 204 | + |
| 205 | +* A counter you base logic on (eg, +slots_taken+) |
| 206 | +* A counter users see (eg, +current_students+) |
| 207 | + |
| 208 | +The reason you want two counters is you'll need to change the value of the logic |
| 209 | +counter *first*, _before_ checking it, to address any race conditions. This means |
| 210 | +the value can get wonky momentarily (eg, there could be 32 +slots_taken+ for a 30-person |
| 211 | +course). This doesn't affect its function - indeed, it's part of what makes it work - but |
| 212 | +does mean you don't want to display it. |
| 213 | + |
| 214 | +So, taking our +Course+ example: |
| 215 | + |
| 216 | + class Course < ActiveRecord::Base |
| 217 | + include Redis::Atoms |
| 218 | + |
| 219 | + counter :slots_taken |
| 220 | + counter :current_students |
| 221 | + end |
| 222 | + |
| 223 | +Then: |
| 224 | + |
| 225 | + @course = Course.find(1) |
| 226 | + @course.slots_taken.increment do |val| |
| 227 | + if val <= @course.max_students |
| 228 | + @course.course_students.create!(:student_id => 101) |
| 229 | + @course.current_students.increment |
| 230 | + end |
| 231 | + end |
| 232 | + |
| 233 | +Race-condition free. And, with the separate +current_students+ counter, your |
| 234 | +views get consistent information about the course, since it will only be |
| 235 | +incremented on success. There is still a race condition where +current_students+ |
| 236 | +could be less than the real number of +CourseStudent+ records, but since you'll be |
| 237 | +displaying these values in a view (after that block completes) you shouldn't see |
| 238 | +this manifest in real-world usage. |
| 239 | + |
| 240 | +Now you can sleep soundly, without fear of getting fired at 3am via an angry |
| 241 | +phone call from your boss. (At least, not about this...) |
| 242 | + |
| 243 | +== Author |
| 244 | + |
| 245 | +Copyright (c) 2009 {Nate Wiger}[http://nate.wiger.org]. All Rights Reserved. |
| 246 | +Released under the {Artistic License}[http://www.opensource.org/licenses/artistic-license-2.0.php]. |
0 commit comments