Unnecessary methods were also removed from CachePT and CacheMT
as it does not make sense to create more than one single instance
of those per filter instance. Consequently there is no need for
them to be able to use an existing StorageFactory (and CacheRules).
The maximum count and maximum size of the cache can now be
specified and a storage can declare what capabilities it has.
If a storage modile cannot enforce the maximum count or maximum
size limits, the storage is decorated with an LRU storage that
can.
The LRUStorage hierarchy implements the decorator pattern and
is capable of providing LRU behaviour to any underlying storage.
There are two concrete classes - LRUStorageST and LRUStorageMT -
that can be used in single thread and multiple thread contentx,
respectively.
Also tests the convention of placing _ as suffix on member
variables.
The query throttling functionality was broken by the move to the per
thread rule system. One way to fix this is to move the session specific
throttling information into the client session itself. This allows for a
simpler system with no direct dependencies on the rules or users
themselves.
Storage is now abstract with StorageReal (that uses an actual
storage module) derived from it. This to prepare for a stackable
case where LRU behaviour is implemented in front of the actual
physcal storage.
The per thread cache is implemented so that there is one single
thread cache per thread.
This uses shared_ptr, which is available in different places
depending on which GCC version is default on the target platform.
So, before this is merged to develop, that needs to be sorted out.
A simple cache is one that does not transparently use multiple
storages in the background. That will be the case when a separate
cache per thread is used. Both CacheST and CacheMT are now derived
from CacheSimple.
A fair amount of what used to be in Cache has now been moved to
CacheSimple. What remains is what surely is common for all cache
types.
Now the basic structure is in place:
- cachefilter.cc is the MaxScale filter interface.
- Cache is the actual cache class that will also handle LRU issues.
- SessionCache (sessioncache.cc) is the session specific cache class
that using Cache acts as the cache for a particular session.
If an item is stale, then one SessionCache will update it.
- StorageFactory is the component that is capable of loading a module
providing storage facilities.
- Storage is the actual key/value store.
The storage module is abstracted with StorageFactory that is capable
of creating Storage instances. The latter contains the data and
provides the behaviour for using the actual storage implementation,
which sits behing a C API, conveniently.
The server test used the wrong name.
MySQL users test loaded multiple modules in one function call and wasn't
appropriate for an internal test suite test as it requires a working
installation.
The cache filter didn't set the library paths before trying to load
modules.
The binlogrouter was missing a NULL check which caused a crash.
If an item cannot be put to the database, it is explicitly removed
to ensure that we cannot have the situation that a stale item is
continuously returned because the updating of the value fails for
whatever reason.
If a cached entry becomes stale, then if no extra measures are
taken then every thread hitting that item while it is being fetched
from the server will also refresh it, even though it obviously is
sufficient that one does it.
Now the knowledge that the item is being refreshed is recorded,
so that all other clients are simply returned the stale item. That
way, we'll hit the server once per stale item.
The rules and users are stored in thread-local pointers which removes the
need to hold global locks. This allows greater scalability but causes a
slightly larger memory footprint. Usually the increase in memory
consumption is trivial compared to the benefits in scaling.
Using a per-thread rule and user list allows changes to be applied
immediately without locking on the instance level.
The updating of the rules uses the new functinality described in the
debugcmd.h header. The module registers two functions at startup; one for
reloading rules and one for printing them. These custom functions can be
seen in the `maxadmin list functions` output.
Since it's a cache, we do not need to retain the data from one
MaxScale invocation to the next. Consequently, we can turn off
write ahead logging completely if we simply wipe the RocksDB
database at each startup. In addition, the location of the
cache directory can now be specified explicitly, so it can be
placed, for instance, on a RAM disk.
The code for the utility is now stored in a separate file. This also
removes the need to include testing headers from other directories.
Also added a function to reload rules that uses the newly modified rule
parsing mechanism. This can be used later on to update the rules at
runtime.
The object names in the dbfwfilter weren't very descriptive and were not
that easy to comprehend. The same goes for function names which had
somewhat cryptic names.
With the advent of qc_get_field_info, columns can now be matched.
However, there is still some undeterminism caused by the table
information not containing contextual information (exactly where
is the table used).
Further, suppose table X contains the column A and table Y contains
the column B, then given a statement like
SELECT a, b from X, Z;
we cannot know whether a is in X or Z, or b in X or Z, without being
aware of the schema, which we currently are not.
Consequently, as long as MaxScale is not aware of the schema, some
heuristics must be applied. For instance, if exactly one table is
referred to, then we can assume that columns that are not explicitly
qualified are from that table.
The rule tests are currently rather rudimentary and need to be
expanded.