Clean connection state up after protocol named prepared statement #163
Conversation
| let _ = final_buffer.get_i32() as usize; | ||
|
|
||
| // Almost certainly true | ||
| if first_message_code == 'P' { |
There was a problem hiding this comment.
Why not run this check in 'P' => { block? Should be cheaper than cloning the entire buffer (which can be very large) every S (effectively every query).
There was a problem hiding this comment.
Also, aren't all Postgres strings NULL-terminated, so this check would never fire?
There was a problem hiding this comment.
Good point on the clone, the reason I run this in the S and not in the P is that we have two places where we handle P buffering so I am trying to avoid duplication. Let me see if we can get away with no clones.
re: NULL-terminated, we only care about the first byte after P and the 32bit size. If it is \0 the this is an unnamed prepared statement, anything else would be interpreted as a named prepared statement.
|
I made a change to access the buffer without cloning. Using |
|
Pretty cool! |
Avoid reporting ProtocolSyncError when admin session disconnects (postgresml#160) Better logging for failure to get connection from pool (postgresml#161) Send signal even if process is gone (postgresml#162) Clean connection state up after protocol named prepared statement (postgresml#163) Add Discord link (postgresml#164) Add SHOW CLIENTS / SHOW SERVERS + Stats refactor and tests (postgresml#159)
We only send
DISCARD ALLif we created a named prepared statement or issued aSET xxxquery. This PR catches an edge case when we create named prepared statements usingPmessage.