]> git.saurik.com Git - redis.git/blob - src/redis-trib.rb
Introduced three client limit classes: normal, slave, pubsub
[redis.git] / src / redis-trib.rb
1 #!/usr/bin/env ruby
2
3 # TODO (temporary here, we'll move this into the Github issues once
4 # redis-trib initial implementation is complted).
5 #
6 # - Make sure that if the rehashing fails in the middle redis-trib will try
7 # to recover.
8 # - When redis-trib performs a cluster check, if it detects a slot move in
9 # progress it should prompt the user to continue the move from where it
10 # stopped.
11 # - Gracefully handle Ctrl+C in move_slot to prompt the user if really stop
12 # while rehashing, and performing the best cleanup possible if the user
13 # forces the quit.
14 # - When doing "fix" set a global Fix to true, and prompt the user to
15 # fix the problem if automatically fixable every time there is something
16 # to fix. For instance:
17 # 1) If there is a node that pretend to receive a slot, or to migrate a
18 # slot, but has no entries in that slot, fix it.
19 # 2) If there is a node having keys in slots that are not owned by it
20 # fix this condiiton moving the entries in the same node.
21 # 3) Perform more possibly slow tests about the state of the cluster.
22 # 4) When aborted slot migration is detected, fix it.
23
24 require 'rubygems'
25 require 'redis'
26
27 ClusterHashSlots = 4096
28
29 def xputs(s)
30 printf s
31 STDOUT.flush
32 end
33
34 class ClusterNode
35 def initialize(addr)
36 s = addr.split(":")
37 if s.length != 2
38 puts "Invalid node name #{addr}"
39 exit 1
40 end
41 @r = nil
42 @info = {}
43 @info[:host] = s[0]
44 @info[:port] = s[1]
45 @info[:slots] = {}
46 @dirty = false # True if we need to flush slots info into node.
47 @friends = []
48 end
49
50 def friends
51 @friends
52 end
53
54 def slots
55 @info[:slots]
56 end
57
58 def to_s
59 "#{@info[:host]}:#{@info[:port]}"
60 end
61
62 def connect(o={})
63 return if @r
64 xputs "Connecting to node #{self}: "
65 begin
66 @r = Redis.new(:host => @info[:host], :port => @info[:port])
67 @r.ping
68 rescue
69 puts "ERROR"
70 puts "Sorry, can't connect to node #{self}"
71 exit 1 if o[:abort]
72 @r = nil
73 end
74 puts "OK"
75 end
76
77 def assert_cluster
78 info = @r.info
79 if !info["cluster_enabled"] || info["cluster_enabled"].to_i == 0
80 puts "Error: Node #{self} is not configured as a cluster node."
81 exit 1
82 end
83 end
84
85 def assert_empty
86 if !(@r.cluster("info").split("\r\n").index("cluster_known_nodes:1")) ||
87 (@r.info['db0'])
88 puts "Error: Node #{self} is not empty. Either the node already knows other nodes (check with nodes-info) or contains some key in database 0."
89 exit 1
90 end
91 end
92
93 def load_info(o={})
94 self.connect
95 nodes = @r.cluster("nodes").split("\n")
96 nodes.each{|n|
97 # name addr flags role ping_sent ping_recv link_status slots
98 split = n.split
99 name,addr,flags,role,ping_sent,ping_recv,link_status = split[0..6]
100 slots = split[7..-1]
101 info = {
102 :name => name,
103 :addr => addr,
104 :flags => flags.split(","),
105 :role => role,
106 :ping_sent => ping_sent.to_i,
107 :ping_recv => ping_recv.to_i,
108 :link_status => link_status
109 }
110 if info[:flags].index("myself")
111 @info = @info.merge(info)
112 @info[:slots] = {}
113 slots.each{|s|
114 if s[0..0] == '['
115 # Fixme: for now skipping migration entries
116 elsif s.index("-")
117 start,stop = s.split("-")
118 self.add_slots((start.to_i)..(stop.to_i))
119 else
120 self.add_slots((s.to_i)..(s.to_i))
121 end
122 } if slots
123 @dirty = false
124 @r.cluster("info").split("\n").each{|e|
125 k,v=e.split(":")
126 k = k.to_sym
127 v.chop!
128 if k != :cluster_state
129 @info[k] = v.to_i
130 else
131 @info[k] = v
132 end
133 }
134 elsif o[:getfriends]
135 @friends << info
136 end
137 }
138 end
139
140 def add_slots(slots)
141 slots.each{|s|
142 @info[:slots][s] = :new
143 }
144 @dirty = true
145 end
146
147 def flush_node_config
148 return if !@dirty
149 new = []
150 @info[:slots].each{|s,val|
151 if val == :new
152 new << s
153 @info[:slots][s] = true
154 end
155 }
156 @r.cluster("addslots",*new)
157 @dirty = false
158 end
159
160 def info_string
161 # We want to display the hash slots assigned to this node
162 # as ranges, like in: "1-5,8-9,20-25,30"
163 #
164 # Note: this could be easily written without side effects,
165 # we use 'slots' just to split the computation into steps.
166
167 # First step: we want an increasing array of integers
168 # for instance: [1,2,3,4,5,8,9,20,21,22,23,24,25,30]
169 slots = @info[:slots].keys.sort
170
171 # As we want to aggregate adiacent slots we convert all the
172 # slot integers into ranges (with just one element)
173 # So we have something like [1..1,2..2, ... and so forth.
174 slots.map!{|x| x..x}
175
176 # Finally we group ranges with adiacent elements.
177 slots = slots.reduce([]) {|a,b|
178 if !a.empty? && b.first == (a[-1].last)+1
179 a[0..-2] + [(a[-1].first)..(b.last)]
180 else
181 a + [b]
182 end
183 }
184
185 # Now our task is easy, we just convert ranges with just one
186 # element into a number, and a real range into a start-end format.
187 # Finally we join the array using the comma as separator.
188 slots = slots.map{|x|
189 x.count == 1 ? x.first.to_s : "#{x.first}-#{x.last}"
190 }.join(",")
191
192 "[#{@info[:cluster_state].upcase}] #{self.info[:name]} #{self.to_s} slots:#{slots} (#{self.slots.length} slots)"
193 end
194
195 def info
196 @info
197 end
198
199 def is_dirty?
200 @dirty
201 end
202
203 def r
204 @r
205 end
206 end
207
208 class RedisTrib
209 def initialize
210 @nodes = []
211 end
212
213 def check_arity(req_args, num_args)
214 if ((req_args > 0 and num_args != req_args) ||
215 (req_args < 0 and num_args < req_args.abs))
216 puts "Wrong number of arguments for specified sub command"
217 exit 1
218 end
219 end
220
221 def add_node(node)
222 @nodes << node
223 end
224
225 def get_node_by_name(name)
226 @nodes.each{|n|
227 return n if n.info[:name] == name.downcase
228 }
229 return nil
230 end
231
232 def check_cluster
233 puts "Performing Cluster Check (using node #{@nodes[0]})"
234 errors = []
235 show_nodes
236 # Check if all the slots are covered
237 slots = {}
238 @nodes.each{|n|
239 slots = slots.merge(n.slots)
240 }
241 if slots.length == 4096
242 puts "[OK] All 4096 slots covered."
243 else
244 errors << "[ERR] Not all 4096 slots are covered by nodes."
245 puts errors[-1]
246 end
247 return errors
248 end
249
250 def alloc_slots
251 slots_per_node = ClusterHashSlots/@nodes.length
252 i = 0
253 @nodes.each{|n|
254 first = i*slots_per_node
255 last = first+slots_per_node-1
256 last = ClusterHashSlots-1 if i == @nodes.length-1
257 n.add_slots first..last
258 i += 1
259 }
260 end
261
262 def flush_nodes_config
263 @nodes.each{|n|
264 n.flush_node_config
265 }
266 end
267
268 def show_nodes
269 @nodes.each{|n|
270 puts n.info_string
271 }
272 end
273
274 def join_cluster
275 # We use a brute force approach to make sure the node will meet
276 # each other, that is, sending CLUSTER MEET messages to all the nodes
277 # about the very same node.
278 # Thanks to gossip this information should propagate across all the
279 # cluster in a matter of seconds.
280 first = false
281 @nodes.each{|n|
282 if !first then first = n.info; next; end # Skip the first node
283 n.r.cluster("meet",first[:host],first[:port])
284 }
285 end
286
287 def yes_or_die(msg)
288 print "#{msg} (type 'yes' to accept): "
289 STDOUT.flush
290 if !(STDIN.gets.chomp.downcase == "yes")
291 puts "Aborting..."
292 exit 1
293 end
294 end
295
296 def load_cluster_info_from_node(nodeaddr)
297 node = ClusterNode.new(ARGV[1])
298 node.connect(:abort => true)
299 node.assert_cluster
300 node.load_info(:getfriends => true)
301 add_node(node)
302 node.friends.each{|f|
303 fnode = ClusterNode.new(f[:addr])
304 fnode.connect()
305 fnode.load_info()
306 add_node(fnode)
307 }
308 end
309
310 # Given a list of source nodes return a "resharding plan"
311 # with what slots to move in order to move "numslots" slots to another
312 # instance.
313 def compute_reshard_table(sources,numslots)
314 moved = []
315 # Sort from bigger to smaller instance, for two reasons:
316 # 1) If we take less slots than instanes it is better to start getting from
317 # the biggest instances.
318 # 2) We take one slot more from the first instance in the case of not perfect
319 # divisibility. Like we have 3 nodes and need to get 10 slots, we take
320 # 4 from the first, and 3 from the rest. So the biggest is always the first.
321 sources = sources.sort{|a,b| b.slots.length <=> a.slots.length}
322 source_tot_slots = sources.inject(0) {|sum,source| sum+source.slots.length}
323 sources.each_with_index{|s,i|
324 # Every node will provide a number of slots proportional to the
325 # slots it has assigned.
326 n = (numslots.to_f/source_tot_slots*s.slots.length)
327 if i == 0
328 n = n.ceil
329 else
330 n = n.floor
331 end
332 s.slots.keys.sort[(0...n)].each{|slot|
333 if moved.length < numslots
334 moved << {:source => s, :slot => slot}
335 end
336 }
337 }
338 return moved
339 end
340
341 def show_reshard_table(table)
342 table.each{|e|
343 puts " Moving slot #{e[:slot]} from #{e[:source].info[:name]}"
344 }
345 end
346
347 def move_slot(source,target,slot,o={})
348 # We start marking the slot as importing in the destination node,
349 # and the slot as migrating in the target host. Note that the order of
350 # the operations is important, as otherwise a client may be redirected to
351 # the target node that does not yet know it is importing this slot.
352 print "Moving slot #{slot} from #{source.info_string}: "; STDOUT.flush
353 target.r.cluster("setslot",slot,"importing",source.info[:name])
354 source.r.cluster("setslot",slot,"migrating",source.info[:name])
355 # Migrate all the keys from source to target using the MIGRATE command
356 while true
357 keys = source.r.cluster("getkeysinslot",slot,10)
358 break if keys.length == 0
359 keys.each{|key|
360 source.r.migrate(target.info[:host],target.info[:port],key,0,1)
361 print "." if o[:verbose]
362 STDOUT.flush
363 }
364 end
365 puts
366 # Set the new node as the owner of the slot in all the known nodes.
367 @nodes.each{|n|
368 n.r.cluster("setslot",slot,"node",target.info[:name])
369 }
370 end
371
372 # redis-trib subcommands implementations
373
374 def check_cluster_cmd
375 load_cluster_info_from_node(ARGV[1])
376 check_cluster
377 end
378
379 def reshard_cluster_cmd
380 load_cluster_info_from_node(ARGV[1])
381 errors = check_cluster
382 if errors.length != 0
383 puts "Please fix your cluster problems before resharding."
384 exit 1
385 end
386 numslots = 0
387 while numslots <= 0 or numslots > 4096
388 print "How many slots do you want to move (from 1 to 4096)? "
389 numslots = STDIN.gets.to_i
390 end
391 target = nil
392 while not target
393 print "What is the receiving node ID? "
394 target = get_node_by_name(STDIN.gets.chop)
395 if not target
396 puts "The specified node is not known, please retry."
397 end
398 end
399 sources = []
400 puts "Please enter all the source node IDs."
401 puts " Type 'all' to use all the nodes as source nodes for the hash slots."
402 puts " Type 'done' once you entered all the source nodes IDs."
403 while true
404 print "Source node ##{sources.length+1}:"
405 line = STDIN.gets.chop
406 src = get_node_by_name(line)
407 if line == "done"
408 if sources.length == 0
409 puts "No source nodes given, operation aborted"
410 exit 1
411 else
412 break
413 end
414 elsif line == "all"
415 @nodes.each{|n|
416 next if n.info[:name] == target.info[:name]
417 sources << n
418 }
419 break
420 elsif not src
421 puts "The specified node is not known, please retry."
422 elsif src.info[:name] == target.info[:name]
423 puts "It is not possible to use the target node as source node."
424 else
425 sources << src
426 end
427 end
428 puts "\nReady to move #{numslots} slots."
429 puts " Source nodes:"
430 sources.each{|s| puts " "+s.info_string}
431 puts " Destination node:"
432 puts " #{target.info_string}"
433 reshard_table = compute_reshard_table(sources,numslots)
434 puts " Resharding plan:"
435 show_reshard_table(reshard_table)
436 print "Do you want to proceed with the proposed reshard plan (yes/no)? "
437 yesno = STDIN.gets.chop
438 exit(1) if (yesno != "yes")
439 reshard_table.each{|e|
440 move_slot(e[:source],target,e[:slot],:verbose=>true)
441 }
442 end
443
444 def create_cluster_cmd
445 puts "Creating cluster"
446 ARGV[1..-1].each{|n|
447 node = ClusterNode.new(n)
448 node.connect(:abort => true)
449 node.assert_cluster
450 node.load_info
451 node.assert_empty
452 add_node(node)
453 }
454 puts "Performing hash slots allocation on #{@nodes.length} nodes..."
455 alloc_slots
456 show_nodes
457 yes_or_die "Can I set the above configuration?"
458 flush_nodes_config
459 puts "** Nodes configuration updated"
460 puts "** Sending CLUSTER MEET messages to join the cluster"
461 join_cluster
462 check_cluster
463 end
464 end
465
466 COMMANDS={
467 "create" => ["create_cluster_cmd", -2, "host1:port host2:port ... hostN:port"],
468 "check" => ["check_cluster_cmd", 2, "host:port"],
469 "reshard" => ["reshard_cluster_cmd", 2, "host:port"]
470 }
471
472 # Sanity check
473 if ARGV.length == 0
474 puts "Usage: redis-trib <command> <arguments ...>"
475 puts
476 COMMANDS.each{|k,v|
477 puts " #{k.ljust(20)} #{v[2]}"
478 }
479 puts
480 exit 1
481 end
482
483 rt = RedisTrib.new
484 cmd_spec = COMMANDS[ARGV[0].downcase]
485 if !cmd_spec
486 puts "Unknown redis-trib subcommand '#{ARGV[0]}'"
487 exit 1
488 end
489 rt.check_arity(cmd_spec[1],ARGV.length)
490
491 # Dispatch
492 rt.send(cmd_spec[0])