Yes, the presence of a table should short-circuit further action. Specifically regarding this case, it unquestionably should not be changing this attribute.
Thanks, - Andrew > On Jul 6, 2026, at 9:07 AM, Dávid Paksy <[email protected]> wrote: > > Hi Devs, > > We are observing an issue where `CREATE TABLE IF NOT EXISTS` is not > idempotent regarding underlying HBase physical properties. > > If an existing table is declared again via `CREATE TABLE IF NOT EXISTS` but > with modified table properties (e.g., changing REGION_REPLICATION from 3 to > 2), the Phoenix metadata layer correctly skips updating SYSTEM.CATALOG, but > the physical HBase TableDescriptor is still mutated. > > Steps to reproduce: > 1. Create table with REGION_REPLICATION=3. > 2. Re-run identical statement with IF NOT EXISTS but change > REGION_REPLICATION=2. > 3. Observe via HBase Shell 'describe' that physical replication dropped to > 2. > > (See PHOENIX-7955 for more details.) > > Shouldn't the presence of the table completely short-circuit the execution > path before hitting the HBase Admin descriptor mutation layers? > > What do you all think? > > Many thanks in advance, > Dávid
