overleaf/services/document-updater/app/coffee/ProjectFlusher.coffee
2019-05-02 16:43:35 +01:00

61 lines
No EOL
2.3 KiB
CoffeeScript

request = require("request")
Settings = require('settings-sharelatex')
RedisManager = require("./RedisManager")
rclient = RedisManager.rclient
docUpdaterKeys = Settings.redis.documentupdater.key_schema
async = require("async")
ProjectManager = require("./ProjectManager")
_ = require("lodash")
ProjectFlusher =
# iterate over keys asynchronously using redis scan (non-blocking)
# handle all the cluster nodes or single redis server
_getKeys: (pattern, limit, callback) ->
nodes = rclient.nodes?('master') || [ rclient ];
doKeyLookupForNode = (node, cb) ->
ProjectFlusher._getKeysFromNode node, pattern, limit, cb
async.concatSeries nodes, doKeyLookupForNode, callback
_getKeysFromNode: (node, pattern, limit = 1000, callback) ->
cursor = 0 # redis iterator
keySet = {} # use hash to avoid duplicate results
batchSize = if limit? then Math.min(limit, 1000) else 1000
# scan over all keys looking for pattern
doIteration = (cb) ->
node.scan cursor, "MATCH", pattern, "COUNT", batchSize, (error, reply) ->
return callback(error) if error?
[cursor, keys] = reply
for key in keys
keySet[key] = true
keys = Object.keys(keySet)
noResults = cursor == "0" # redis returns string results not numeric
limitReached = (limit? && keys.length >= limit)
if noResults || limitReached
return callback(null, keys)
else
setTimeout doIteration, 10 # avoid hitting redis too hard
doIteration()
# extract ids from keys like DocsWithHistoryOps:57fd0b1f53a8396d22b2c24b
# or docsInProject:{57fd0b1f53a8396d22b2c24b} (for redis cluster)
_extractIds: (keyList) ->
ids = for key in keyList
m = key.match(/:\{?([0-9a-f]{24})\}?/) # extract object id
m[1]
return ids
flushAllProjects: (limit, concurrency = 5, callback)->
ProjectFlusher._getKeys docUpdaterKeys.docsInProject({project_id:"*"}), limit, (error, project_keys) ->
if error?
logger.err err:error, "error getting keys for flushing"
return callback(error)
project_ids = ProjectFlusher._extractIds(project_keys)
jobs = _.map project_ids, (project_id)->
return (cb)->
ProjectManager.flushAndDeleteProjectWithLocks project_id, cb
async.parallelLimit jobs, concurrency, (error)->
return callback(error, project_ids)
module.exports = ProjectFlusher