| Overview |  |
Behaviours: eysql_driver.
The default driver: epgsql, with the sharp edges filed off.
epgsql:connect/1 links the socket process to the caller before
connecting, and that process exits abnormally when the connect fails, so a
failed connect can kill the caller. open/3 starts the socket
process unlinked and links it only once the connection is up.
An epgsql call on a connection whose process has died exits the caller.
squery/2 and equery/3 turn that into
{error, {connection_lost, Reason}}.
epgsql waits for the server for as long as it takes, and a server that
has died silently or hangs never answers. On a connection checked out of
a pool with a socket_timeout, in the process that checked it out,
squery/2 and equery/3 wait that long at most: the pool
then kills the connection, and they return
{error, {connection_lost, socket_timeout}} (see
eysql_pool:watch/1). So code inside eysql:with_connection/3
or eysql:transaction/3 gets bounded queries by calling them in
place of epgsql:squery/2 and epgsql:equery/3. Anywhere else they
behave as those do, with no bound.
transaction_status/1 reads it inside the connection
process.
transaction_status() = idle | in_transaction | failed | unknown
| close/1 | |
| committed/1 | |
| discover/1 | Read yb_servers(). |
| equery/3 | epgsql:equery/3, as squery/2 is epgsql:squery/2. |
| is_primary/1 | |
| open/3 | Connect to one host. |
| squery/2 | epgsql:squery/2, returning an error rather than exiting when the
connection has died, and bounded by the pool's socket_timeout when this
process checked Conn out of a pool. |
| transaction_status/1 | Whether the server has a transaction open on the connection, from the
status byte of its last ReadyForQuery: idle (I), in_transaction (T) or
failed (E, a statement failed and only ROLLBACK is accepted). |
close(Conn::pid()) -> ok
committed(Conn::pid()) -> boolean()
discover(Conn::pid()) -> {ok, [eysql_topology:server()]} | {error, term()}
Read yb_servers(). On PostgreSQL this fails with SQLSTATE 42883
(undefined function), which the cluster takes to mean "not YugabyteDB".
public_ip is empty for a server that has none.
equery(Conn::pid(), Sql::iodata(), Params::[term()]) -> term()
epgsql:equery/3, as squery/2 is epgsql:squery/2.
is_primary(Conn::pid()) -> {ok, boolean()} | {error, term()}
open(Host::binary(), Port::inet:port_number(), Settings::map()) -> {ok, pid()} | {error, term()}
Connect to one host. The connection is linked to the caller, as with
epgsql:connect/1.
squery(Conn::pid(), Sql::iodata()) -> term()
epgsql:squery/2, returning an error rather than exiting when the
connection has died, and bounded by the pool's socket_timeout when this
process checked Conn out of a pool.
transaction_status(Conn::pid()) -> transaction_status()
Whether the server has a transaction open on the connection, from the
status byte of its last ReadyForQuery: idle (I), in_transaction (T) or
failed (E, a statement failed and only ROLLBACK is accepted). No round
trip to the server.
unknown when the byte does not tell: the process is gone or took longer
than a second to answer, a command is still running, epgsql waits for a
sync after an extended-query error, the connection is in COPY mode, or the
state is not one this function knows.
After epgsql:parse, epgsql:bind or epgsql:execute with no
epgsql:sync/1 after them, the byte is stale: the server sends
ReadyForQuery only on sync.
Generated by EDoc