This is not exactly correct. When you do an SQL query with only PARTITIONED tables, or with a mix of PARTITIONED and REPLICATED, the data will be taken from the primary partions of PARTITIONED tables and *all* partitions of REPLICATED tables. When you do an SQL query with only REPLICATED tables, the data will be taken from all partitions (as they're all always available).
E.g. take this query SELECT * FROM A JOIN B WHERE A.FIELD_A = B.FIELD_B This query will work correctly if the data is collocated, i.e. in the following cases: - A and B are both REPLICATED - A is PARTIONED and B is REPLICATED - A is REPLICATED and B is PARTITIONED - A and B are both PARTITIONED with FIELD_A and FIELD_B both being their affinity keys If A and B are both PARTITIONED but FIELD_A or FIELD_B are not affinity keys then the data is not correctly collocated and the query will return incorrect results. Stan Stan On Mon, Oct 28, 2019 at 9:34 PM <alexanderko...@gmail.com> wrote: > In PARTITIONED mode *SQL queries are executed* over each node’s *primary* > partitions only. > > > > Here is more information about distributed joins: > https://apacheignite-sql.readme.io/docs/distributed-joins > > > > > > > > *From:* Prasad Bhalerao <prasadbhalerao1...@gmail.com> > *Sent:* Saturday, October 26, 2019 12:25 AM > *To:* user@ignite.apache.org > *Subject:* Re: Query execution in ignite > > > > > > Question is specifically about primary and secondary partitions. > > > > So in case of replicated cache ignite scans primary and secondary > partitions of any one of the node of the cluster to to fetch the data. > > > > But is it the case with partitioned cache. > > I mean in case of partitioned cache when SQL is executed, does ignite scan > primary as well as secondary partitions of each node in the cluster or it > just scans primary partitions of all the nodes in the cluster as the query > is being executed on all nodes? > > > > On Fri 25 Oct, 2019, 10:46 PM <alexanderko...@gmail.com wrote: > > Hi, > > > > If a query is executed against a fully *REPLICATED* data then Ignite > will send it to a single cluster node and run it over the local data there. > > > > > > > > if a query is executed over a *PARTITIONED* data, then the execution > flow will be the following: > > The query will be parsed and split into multiple map queries and a single > reduce query. > > · All the map queries are executed on all the nodes where > required data resides. > > · All the nodes provide result sets of local execution to the > query initiator (reducer) that, in turn, will accomplish the reduce phase > by properly merging provided result sets. > > > > > > > > More information here: > https://apacheignite-sql.readme.io/docs/how-ignite-sql-works > > Thanks, Alex > > > > *From:* Prasad Bhalerao <prasadbhalerao1...@gmail.com> > *Sent:* Friday, October 25, 2019 1:31 AM > *To:* user@ignite.apache.org > *Subject:* Query execution in ignite > > > > Hi, > > > > When SQL is executed, does ignite always scan only primary partitions of > all available nodes in cluster irrespective of cache mode partitioned or > replicated? > > > > > > > > Thanks , > > Prasad > >
