kunal642 commented on a change in pull request #3294: [CARBONDATA-3462][DOC]Added documentation for index server URL: https://github.com/apache/carbondata/pull/3294#discussion_r303731159
########## File path: docs/index-server.md ########## @@ -0,0 +1,238 @@ +<!-- + Licensed to the Apache Software Foundation (ASF) under one or more + contributor license agreements. See the NOTICE file distributed with + this work for additional information regarding copyright ownership. + The ASF licenses this file to you under the Apache License, Version 2.0 + (the "License"); you may not use this file except in compliance with + the License. You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. +--> + +# Distributed Index Server + +## Background + +Carbon currently caches all block/blocklet datamap index information into the driver. For bloom +datamap, it can prune the splits in a distributed way. In the first case, there are limitations +like driver memory scale up and cache sharing between multiple applications is not possible. In +the second case, there are limitations like, there is +no guarantee that the next query goes to the same executor to reuse the cache and hence cache +would be duplicated in multiple executors. +Distributed Index Cache Server aims to solve the above mentioned problems. + +## Distribution +When enabled, any query on a carbon table will be routed to the index server application using +the Hadoop RPC framework in form of a request. The request will consist of the table name, segments, +filter expression and other information used for pruning. + +In IndexServer application a pruning RDD is fired which will take care of the pruning for that +request. This RDD will be creating tasks based on the number of segments that are applicable for +pruning. It can happen that the user has specified segments to access for that table, so only the +specified segments would be applicable for pruning. + +IndexServer driver would have 2 important tasks, distributing the segments equally among the +available executors and keeping track of the cache location(where the segment cache is present). + +To achieve this 2 separate mappings would be maintained as follows. +1. segment to executor location: +This mapping will be maintained for each table and will enable the index server to track the +cache location for each segment. +``` +tableToExecutorMapping = Map(tableName -> Map(segmentNo -> uniqueExecutorIdentifier)) +``` +2. Cache size held by each executor: + This mapping will be used to distribute the segments equally(on the basis of size) among the + executors. +``` +executorToCacheMapping = Map(HostAddress -> Map(ExecutorId -> cacheSize)) +``` + +Once a request is received each segment would be iterated over and +checked against tableToExecutorMapping to find if a executor is already +assigned. If a mapping already exists then it means that most +probably(if not evicted by LRU) the segment is already cached in that +executor and the task for that segment has to be fired on this executor. + +If mapping is not found then first check executorToCacheMapping against +the available executor list to find if any unassigned executor is +present and use that executor for the current segment. If all the +executors are assigned with some segment then find the least loaded +executor on the basis of size. + +Initially the segment index size would be used to distribute the +segments fairly among the executor because the actual cache size would +be known to the driver only when the segments are cached and appropriate +information is returned to the driver. + +**NOTE:** In case of legacy segment the index size if not available +therefore all the legacy segments would be processed in a round robin +fashion. + +After the job is completed the tasks would return the cache size held by +each executor which would be updated to the executorToCacheMapping and +the pruned blocklets which would be further used for result fetching. + +## Reallocation of executor +In case executor(s) become dead/unavailable then the segments that were +earlier being handled by those would be reassigned to some other +executor using the distribution logic. Review comment: distribution logic is already mentioned above ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
