Thank Godfrey for driving. Looks very good~ This will undoubtedly greatly enhance the various batch mode connectors.
I left some comments: ## FileBasedStatisticsReportableDecodingFormat One concern I have is that we read the footer for each file, and this may be a bit costly in some cases. Is it possible for us to have some hierarchical way, e.g. - No statistics are collected for files by default. - SIZE: Generate statistics based on file Size, get the size of the file only with access to the master of the FileSystem. - DETAILED: Get the complete statistics by format, possibly by accessing the footer of the file. ## When use the statistics reported by connector > When partitions are pruned by PushPartitionIntoTableSourceScanRule, the statistics should also be updated. I understand that we definitely need to use reporter after the partition prune, but another question: is the filter pushdown already happening at this time? Can we make sure that in the following three cases, both the filter pushdown and the partition prune happen before the stats reporting. - only partition prune happens - only filter pushdown happens - both filter pushdown and partition prune happen Best, Jingsong On Fri, May 13, 2022 at 6:57 PM godfrey he <godfre...@gmail.com> wrote: > Hi all, > > I would like to open a discussion on FLIP-231: Introduce > SupportStatisticReport > to support reporting statistics from source connectors. > > Statistics are one of the most important inputs to the optimizer. > Accurate and complete statistics allows the optimizer to be more powerful. > Currently, the statistics of Flink SQL come from Catalog only, > while many Connectors have the ability to provide statistics, e.g. > FileSystem. > In production, we find many tables in Catalog do not have any statistics. > As a result, the optimizer can't generate better execution plans, > especially for Batch jobs. > > There are two approaches to enhance statistics for the planner, > one is to introduce the "ANALYZE TABLE" syntax which will write > the analyzed result to the catalog, another is to introduce a new > connector interface > which allows the connector itself to report statistics directly to the > planner. > The second one is a supplement to the catalog statistics. > > Here, we will discuss the second approach. Compared to the first one, > the second one is to get statistics in real time, no need to run an > analysis job for each table. This could help improve the user > experience. > (We will also introduce the "ANALYZE TABLE" syntax in other FLIP.) > > You can find more details in FLIP-231 document[1]. Looking forward to > your feedback. > > [1] > https://cwiki.apache.org/confluence/pages/resumedraft.action?draftId=211883860&draftShareId=eda17eaa-43f9-4dc1-9a7d-3a9b5a4bae00&; > [2] POC: https://github.com/godfreyhe/flink/tree/FLIP-231 > > > Best, > Godfrey >
