Hi Nico, Judging by that snippet above, it looks like you're actually using Spark's native JDBC integration though the regular Phoenix driver to create your source dataframe, not the spark-phoenix integration at all [1].
If that's sufficient for you, I suspect you can also use the DataFrame JDBC write() method, as per [2]. FWIW, internally when using the phoenix-spark integration, we just pass the tenantID manually as part of the predicate, since our analytics layer is aware of all tenants. Good luck, Josh [1] https://phoenix.apache.org/phoenix_spark.html [2] http://stackoverflow.com/a/32625114 On Fri, Oct 7, 2016 at 1:14 PM, James Taylor <[email protected]> wrote: > Not sure if the phoenix-spark integration propagates connection properties, > but an alternate way of specifying the tenant ID is through that mechanism > (which would workaround the URL building issue). > > On Fri, Oct 7, 2016 at 10:06 AM, Nico Pappagianis < > [email protected]> wrote: > > > Hi James, > > At first I thought it was the dataframe vs rdd implementations but > looking > > closer my bet is the way spark connects to phoenix. When reading via > > SqlContext I pass in > > > > "url" -> "jdbc:phoenix:zkHost1, zkHost2, zkHost3:zkPort;TenantId= > > 123456789" > > and it connects as the tenant. > > > > However SqlContext does not have a write/save function. > > > > When I try to save by other means I am required to pass in a value for > > "zkUrl" (not "url"). "zkUrl" cannot have the "jdbc:phoenix:" portion > > attached (because it attaches zkPort to the end of jdbc:phoenix and > errors > > out). As such I cannot connect as the tenant. > > > > When connecting as the tenant via squirrel client I use the same "url" > > string above, and it works. > > > > So to me it appears to be an issue of how to connect to phoenix as the > > tenant via spark/phoenix-spark integration. I have not found a clear cut > > way to do so. > > > > > > Thanks, > > -Nico > > > > > > On Fri, Oct 7, 2016 at 9:03 AM, James Taylor <[email protected]> > > wrote: > > > > > Hi Nico, > > > You mentioned offline that it seems to be working for data frames, but > > not > > > RDDs. Can you elaborate on that? Have you confirmed whether the > TenantId > > > connection property is being propagated down to the Phoenix connection > > > opened for the Spark integration? > > > Thanks, > > > James > > > > > > On Thu, Oct 6, 2016 at 8:36 PM, Nico Pappagianis < > > > [email protected]> wrote: > > > > > > > Does phoenix-spark integration support multitenancy? I'm having a > hard > > > time > > > > getting it working on my tenant-specific view. > > > > > > > > Thanks > > > > > > > > > >
