Let's think back to before transactions where added to Phoenix.
With autoCommit=false, updates to HBase will be batched in the Phoenix
driver, eventually flushing on their own or whenever you invoked
commit() on the connection.
With autoCommit=true, updates to HBase are flushed with every executed
statement.
For what you're describing here, not being able to see the table
contents via Squirrel makes sense to make. I don't believe the
introduction of transactions to Phoenix would be intended to change
these semantics.
(QueryServer in the mix also doesn't affect any of this).
Does that make sense?
F21 wrote:
I am using HBase 1.1.3 with Phoenix 4.8.0-SNAPSHOT. To talk to phoenix,
I am using the phoenix query server with serialization set to JSON.
First, I create a non-transactional table:
CREATE TABLE my_table3 (k BIGINT PRIMARY KEY, v VARCHAR)
TRANSACTIONAL=false;
I then send the following requests to the query server using curl:
curl localhost:8765 -XPOST --data '{"request":
"openConnection","connectionId": "my-conn"}'
curl localhost:8765 -XPOST --data '{"request":
"connectionSync","connectionId": "my-conn","connProps": {"connProps":
"connPropsImpl","autoCommit": false,"transactionIsolation": 8}}'
curl localhost:8765 -XPOST --data '{"request":
"createStatement","connectionId": "my-conn"}'
curl localhost:8765 -XPOST --data '{"request":
"prepareAndExecute","connectionId": "my-conn","statementId":
12345,"sql": "UPSERT INTO my_table3 VALUES (1,\'A\')","maxRowCount":
100}' #Update the statement id
curl localhost:8765 -XPOST --data '{"request":
"createStatement","connectionId": "my-conn"}'
curl localhost:8765 -XPOST --data '{"request":
"prepareAndExecute","connectionId": "my-conn","statementId":
12345,"sql": "UPSERT INTO my_table3 VALUES (2,\'B\')","maxRowCount":
100}' #Update the statement id
Connect to the phoenix query server using Squirrel SQL and see no rows
exist for the my_table3 table.
curl localhost:8765 -XPOST --data '{"request":
"createStatement","connectionId": "my-conn"}'
curl localhost:8765 -XPOST --data '{"request":
"prepareAndExecute","connectionId": "my-conn","statementId":
12345,"sql": "SELECT * FROM my_table3","maxRowCount": 100}' # Shows no
results
curl localhost:8765 -XPOST --data '{"request": "commit","connectionId":
"my-conn"}'
Connect to the phoenix query server using Squirrel SQL and see 2 rows
for the my_table3 table.
It seems a bit strange that the table exhibits some properties where it
supports transactions to some extent. Is this something that should be
fixed?
Ideally, if a connectionSync request turns off autocommit for a given
connection, reading and writing to a non-transactional table using that
connection should return an error.
