On Tue, Mar 16, 2021 at 4:28 AM Andres Freund <and...@anarazel.de> wrote:
Replying to some of the comments.. > - Is nodeModifyTable.c really the right place for the logic around > CompareCompressionMethodAndDecompress()? And is doing it in every > place that does "user initiated" inserts really the right way? Why > isn't this done on the tuptoasting level? I think if we do in tuptoasting level then it will be even costlier because in nodeModifyTable.c at least many time we will get the virtual tuple e.g. if a user is directly inserting the tuple but once we go down to tuptoasting level by then we will always get the HeapTuple and we will have to deform in every case where tupdesc has any varlena because we don't have any flag in the tuple header to tell us whether there are any compressed data or not. In the below thread[1] we have considered these 2 approaches and basically, in unrelated paths like pg_bench, we did not see any performance regression with any of those approaches. [1] https://www.postgresql.org/message-id/CAFiTN-vcbfy5ScKVUp16c1N_wzP0RL6EkPBAg_Jm3eDK0ftO5Q%40mail.gmail.com > > I'm open to being convinced that we don't need to do either of these > > things, and that the cost of iterating over all varlenas in the tuple > > is not so bad as to preclude doing things as you have them here. But, > > I'm afraid it's going to be too expensive. > > I mean, I would just define several of those places away by not caring > about tuples in a different compressino formation ending up in a > table... I am just wondering that why we don't need to process in case of storage change, I mean if the target table has the attribute storage as external and if there are some compressed data coming from the source table then we will be inserting those compressed data as it is in the target attribute without externalizing. Maybe it is done to avoid such performance impacts? Well, we can do the same for the compression also and just provide some mechanism to recompress maybe in vacuum full/cluster. -- Regards, Dilip Kumar EnterpriseDB: http://www.enterprisedb.com
