The automatic reconnection feature of the Connector-C was not
enabled. Enabling it should reduce the amount of false positives that the
monitors pick up.
If the session id is known, it will be logged together with all
messages. If present, the session id appears, enclosed in paranthesis,
right after the message category. E.g.
2017-08-30 12:20:49 warning: (4711) [masking] The rule ...
Conceptually this is a cherry-pick of commit
67efd1daeabbc398b8a8fbc0cd02c2af26e4cb6c (2.2), but too much has
changed to actually be able to cherry-pick that commit.
When an unexpected response to a COM_CHANGE_USER is received, it is now
processes and discarded instead of treated as an error. This will allow
further analysis of the situation in addition to possibly solving some of
the problems that the persistent connections have.
Added extra info level logging to relevant parts of the code that deal
with the COM_CHANGE_USER reply processing. This information should allow
tracking of the response state for debugging purposes.
The response to the COM_CHANGE_USER should always be turned into a
contiguous buffer of complete packets. This guarantees that the code that
processes it functions properly.
As COM_QUIT would terminate the connection, there's no need to initiate
the session reset process. Also make sure all buffers are empty before
putting the DCB into the pool.
Added extra debug assertions for parts of the code that are related to the
COM_CHANGE_USER processing.
The function that printed all sessions assumed that all client DCBs had
valid, non-dummy sessions. It is possible that a client with a dummy
session is the list. These sessions should be ignored.
When a persistent connection is reused, a COM_CHANGE_USER command is
executed to reset the session state. If the reused connection was closed
before the response to the COM_CHANGE_USER was received and taken into use
by another connection, another COM_CHANGE_USER would be sent to, again,
reset the session state. Due to the fact that the first response is still
on its way, it will appear as if two responses are generated for a single
COM_CHANGE_USER.
The way to fix this is to avoid putting connections that haven't been
successfully reset into the connection pool.
When a session is being closed in a controlled manner, i.e. a COM_QUIT is
received from the client, it is possible to deduce from this fact that the
backend connections are very likely to be idle. This can be used as an
additional qualification that must be met by all connections before they
can be candidates for connection pooling.
This assumption will not hold with batched and asynchronous queries. In
this case it is possible that the COM_QUIT is received from the client
before even the first result from the backend is read. For this to work,
the protocol module would need to track the number and state of expected
responses.
The buffer used to store the hexadecimal string was one byte too
short. This caused the trailing null terminator to be written into
unallocated memory.
The pointer manipulation in modutil_count_statements assumed that if a
semicolon is found, it is not the last character in the buffer. It also
assumed that the buffer contained at least one readable character.
If the binlog has binlog checksums enabled, the extra checksum bytes are
removed from the end of the event. The avrorouter assumes that whatever
caused the binlogs to appear in the first place already checked that the
checksums are OK.
Also removed one extra byte being added to the length of all query events.
A new option ‘slave_hostname’ allows the setting of hostname in
COM_REGISTER_SLAVE.
SHOW SLAVES HOSTS; in master server can show the hostname set in binlog
router:
MariaDB [(none)]> SHOW SLAVE HOSTS;
+-----------+-----------------------------+------+-----------+
| Server_id | Host | Port | Master_id |
+-----------+-----------------------------+------+-----------+
| 93 | maxscale-blr-1.mydomain.net | 8808 | 10124 |
+-----------+-----------------------------+------+-----------+
If a connection has not been fully established (i.e. authentication has
been completed) then it should not be considered as a connection pool
candidate.
When a buffer is cloned and then the original buffer parsed and freed, the
freeing of the cloned buffer will not release the memory that was
allocated when the original buffer is parsed.
This is a side-effect of how the buffer objects are stored in the buffer
and not in the shared memory buffer. The creation of a buffer object after
cloning will cause the buffer object to be lost as the cloned buffer
didn't have a pointer to the buffer object that was created later.
By moving the buffer objects into the shared memory buffer, the memory
leak is fixed.
Enabling the option hinders the use of maintenance mode with the root
master node in most use-cases.
This behavior occurs due to the fact that the maintenance mode causes a
server to be treted as if it was down. The Galera monitor waits for the
cluster to reorganize before assigning a new master node. This is correct
(but very unexpected) behavior for single instance use-cases.
The query classifier should only be used to parse text protocol
statements. The insertstream filter exploited the fact that any statements
that the filter did not expect would be classified as an unknown
commands. This led to repetitive error messages with binary protocol
statements.
When a backend is waiting for a response but no statement is stored for
the session, the buffer where the stored statement is copied is not
modified. This means that it needs to be initialized to a NULL value.
Added a test that checks that the behavior works as expected even with
persistent connections. A second test reproduces the crash by executing
parallel SET commands while slaves are blocked.
There is still a behavioral problem in readwritesplit. If a session
command is being executed and it fails on a slave, an error is sent to the
client. In this case it would not be necessary to close the session if the
master is still alive.
When an internal connection is created, the SQL_MODE of the connection
should be set to a known default. The empty SQL_MODE allows consistent
functionality for all backend server versions.
The asserted value can be false without it being an error. When a table is
re-mapped to a different position, there is no guarantee that the previous
value has not been reused by another table.
If a rule is defined with only an optional part, it should be of the
permission type. This type is used to signal that the rule matches if the
optional constraints are fulfilled.
Due to refactoring, the default type was changed from RT_PERMISSION to
RT_UNDEFINED.
The schemarouter can now resolve database mapping conflicts in a
deterministic manner. This will fix the problem of central databases which
are replicated shards being assigned in a non-deterministic manner.
When the current database is implicitly used in a query that also uses an
explicit database, it must be routed to the shard which has the current
database.
As cross-shard joins are not supported, the safest, and possibly the most
expected course of action to take, is to route it to the so-called default
shard. The default shard is the shard that contains the database that is
currently set as the active database with a COM_INIT_DB, a text protocol
USE <database> query or it was set at connection time.
The avrorouter failed to detect ALTER TABLE statements which caused a
regression. Extended the alter table tests to parse the JSON for more
strict validation of test results.
The multimaster node detection uses stacks to sort the node groups. The
size of this stack was always assumed to be positive but it was possible
that it dropped down to -1 causing a crash when the stack was accessed
with the index number.
The EVP_CIPHER_CTX is now created inside a wrapper function to add support
for OpenSSL 1.1. Also fixed improper use of the EVP_CIPHER_CTX internals
in binlogrouter.
OpenSSL 1.1 supports most of the native threading libraries, including
pthread. This means that only versions before 1.1 need the thread handling
code.
