Toru Maesaka

Like most things, I think storage engine development is about divide and conquer. The first sub-problem that I’m tackling with BlitzDB is squeezing as much juice out as possible from Tokyo Cabinet to achieve fast write performance. This by the way happens to be the primary reason that I wrote skyload.

Writing skyload turned out to be worthwhile since it helped me find several critical bugs in the engine that only occurred under concurrent insertion load. Thanks to Kazuho Oku for helping me through the issues that I was facing.

I think I’ve now reached a stage where I can share how well BlitzDB can perform insertion from concurrent connections. But before moving ahead, I’d like to emphasize that for a real guideline, I believe that performance comparison should be done by an unbiased third party. So please don’t take the results in this post as the “truth”. Heh, I did write both the storage engine and the load emulator after all :)

So, with the above in mind, here’s a skyload result on inserting one-hundred-thousand rows under different concurrency levels with BlitzDB and MyISAM (both engines under default configuration).

Figures presented above are calculated from an average of 5 runs per each concurrency level. Admittedly, an average from 5 runs is not sufficient to claim credibility of my result since the figures can easily be affected by the dirty buffer flush between the kernel and the filesystem (ext3 in this particular benchmark). For this, I plan on extending skyload to run multiple runs of an identical test and compute the median and average.

For those that are interested, this is what the test table looks like:

CREATE TABLE t1 (
    id int PRIMARY KEY,
    col1 int,
    col2 double,
    col3 varchar(255)
) ENGINE=blitz;

I didn’t bother benchmarking anything beyond 32 connections since I ran both the client and the server on the same quad core machine (there’s no point). This is probably why you can see a nice curve up to four concurrent connections with BlitzDB in the graph. Yet another reason why you should not believe everything I’ve provided in this entry.

BlitzDB needs you

BlitzDB is still very early in it’s making and I still have insane amount of work to do. For example, BlitzDB currently requires you to supply a primary key on your table. I plan on removing this limitation by generating a “fake” primary key internally but I still haven’t got around to it at this point.

Support for multiple indexes is not done yet despite having all the necessary components to achieve it. I could do all this on my own but I prefer not to. I’m totally open for ideas and contributors. If you’re interested in this storage engine project, please don’t hesitate to ping me (dev @ this domain) or the Drizzle community. More eyeballs the merrier :)

Toru Maesaka drizzle, oss blitzdb, storage, tc

I’m playing a little behind but I’d like to spread the word that memcached-1.4.0 has been released.

• http://code.google.com/p/memcached/downloads/list
• http://code.google.com/p/memcached/wiki/ReleaseNotes140

You can also find blog posts by other memcached developers floating around on the internet. This 1.4 release is quite special for me too since the binary protocol is something I’ve been eager to see out in the open for sometime now. I remember printing out the early binary protocol specification (the draft that was made before I joined the community) and reading it on the flight to SFO back in 2007.

Much has happened since then with many input from both developers and non-developers. Several commercial organizations from small to large has helped us evaluate the old experimental branch(es) that we’ve been working on at github. The 1.4 release is a result of community effort that I’m really proud to be part of.

I recommend all of you to update to 1.4 and also take a look at the latest version of your client library. See if it supports the binary protocol. If it doesn’t, you should bug the author about it ;)

Which reminds me, I need to ping Dormando and Hachi about the binary protocol patch that I wrote for Cache::Memcached last year…

Toru Maesaka memcached, oss memcached, release

Yesterday I merged the BlitzDB tree with Drizzle‘s trunk for the first time in a long time (yeah…) and discovered some interesting changes made to the storage subsystem while I was away.

Previously all functions that caused an action to the storage engine was a member of the handler class but various things like table creation and transaction related functions have now moved to the StorageEngine class. These changes are somewhat drastic but makes good sense for Drizzle to grow further since it makes the subsystem easier to understand and frees Drizzle from the interface design that was strongly affected by MyISAM. For those that are interested, the StorageEngine class is located in “drizzled/plugin/storage_engine.h”.

For me it was pretty easy to update BlitzDB to work with the new subsystem since I don’t have anything special in the engine that required me to use my brain. I only had to move bas_ext(), table creation and rename functions over to the StorageEngine class and adjust it to the new interface:

int createTableImpl(Session *session, const char *table_name, 
                    Table *table_arg, HA_CREATE_INFO *ha_create_info); 
 
int renameTableImpl(Session *session, const char *from, const char *to);

For a real example, I recommend comparing the old InnobaseEngine class declaration with the updated one. As for where this redesign is going, this is the answer I got on the Drizzle channel from Stewart who did the actual work for all this.

stewart: tmaesaka: the basic idea is that handler becomes a cursor. the StorageEngine is for actions on the engine.
stewart: tmaesaka: and handler is a cursor on a table.

Something to keep in mind if you’re thinking about creating or porting a storage engine to Drizzle :)

Toru Maesaka drizzle, oss drizzle, mysql, storage

Today, I would like to introduce “skyload“, a small project that I’ve been working on for the last couple of weeks. In brief, skyload is a libdrizzle based load emulation tool that is capable of running concurrent load tests against database instances that can speak Drizzle (and/or) the MySQL protocol.

• http://code.google.com/p/skyload/

Something I’d like to emphasize here is that, skyload is not a replacement for mysqlslap or drizzleslap since it only provides a subset of what they can do. As I’ve stated on the project description, skyload is designed to do a good job at this subset of tasks by giving you more control over how you emulate the load in an intuitive way. For instructions on installing skyload and quickly getting up to speed, take a look at the following URL:

• http://code.google.com/p/skyload/wiki/Tutorial_en

As you will see, the first release only contains bare minimum specifications (only INSERT load emulation). The next step I want to take is to discuss features that other storage engine developers would actually find useful. This is because I started writing skyload for primarily myself and other storage engine developers (more on this next).

Original Intentions

I originally began writing skyload for BlitzDB development since I wanted to see the concurrent insertion performance of Tokyo Cabinet based row storage mechanism that I wrote. I first tried benchmarking the write performance with drizzleslap but it turned out that drizzleslap’s original code (inherited from MySQL 6.0) is rather buggy and segfaulted quite easily (I’m planning on contributing a fix for this).

So I gave up on drizzleslap for the time being and started looking at the sysbench port for Drizzle that Monty Taylor has been working on:

• sysbench: now with Drizzle

Sysbench for Drizzle is a lovely piece of software but it couldn’t quite provide what I was looking for (concurrent insertion benchmark). After having a quick conversation with Monty about my requirements on the Drizzle IRC channel, I decided to write a libdrizzle based benchmark tool that can be used for both Drizzle and MySQL.

Future Plans

I don’t want to reinvent existing software that works (or those that can be fixed). The project positioning that I’m hoping for skyload is a good mix between (mysql|drizzle)slap and sysbench. Hopefully it will be useful to folks that works on Drizzle and MySQL related projects.

I’m totally open for ideas, patches, and contributors. If this project had caught your attention, please don’t hesitate to ping me or the Drizzle community :)

I haven’t setup a mailing list since I don’t see the need for it yet so if you’d like to share your thoughts I think either the Drizzle mailing list or IRC (#drizzle @ irc.freenode.net) is the quickest way for me to get back to you.

Happy Hacking!

Toru Maesaka drizzle, oss benchmark, drizzle, mysql, oss, skyload, tools

Something I’ve added to BlitzDB recently that was pretty high on my todo list is support for variable width tables. So what is a variable width table? it is a table that contains columns that can vary in size, namely BLOB and TEXT types.

Going back to the basics, when a new row is to be written, a storage engine is given a pointer to the row data in MySQL format that it must somehow store for later lookup/retrieval. By meaning “somehow”, the storage engine is given the freedom to do whatever it likes with the row.

Writing a row for a fixed length table (a table with columns that are always the same size) is deadly easy. A storage engine can choose to not temper with the row and simply write or copy the data to it’s storage mechanism. This is because the storage engine is given a row that contains all the data. Rows for variable width tables however, are treated differently since things aren’t as simple (it’s variable!).

The difference is that columns for BLOB and TEXT types are represented by two parts inside a MySQL/Drizzle row:

• length of the data
• pointer to the actual data

This is simple to understand since we need to know the size of the data to copy it.

Minor Complication

The minor complication as you would expect here is that you can’t directly write the provided row to your engine like you can with fixed length tables. The data that you want to copy/write exists elsewhere (hence the pointer) so directly writing the row has no meaning (the data would have disappeared by your next access to that row). You need to make sure that the actual data for BLOB/TEXT column(s) are arranged appropriately on your engine’s row buffer and written out to it’s storage mechanism.

This process is commonly referred to as row packing (converting to your engine format) and unpacking (convert back to MySQL format). So how is this done? it’s actually pretty simple!

The solution is actually simple

As much as it sounds like a bother to support variable length rows, it’s actually not that bad. First you need to understand what a MySQL row looks like internally.

A MySQL row begins with a bitset that represents which fields are NULL. The length of this data obviously depends on the number of NULLable columns you have but this is easy to handle with Drizzle since we’re given all the relevant information by the TableShare object (same goes for MySQL from a different object).

After this data comes the actual column data in the order that appears in your CREATE TABLE statement. What you need to do to get packing working with this row is the not-so-obvious part that you really need an example to look at. Fortunately Tweeting about this attracted Brian’s attention which helped me move forward.

Loop the fields!

So, let’s take row insertion to a variable width table as an example. Imagine this table:

CREATE TABLE t1 (
  id int PRIMARY KEY NOT NULL,
  description text,
  arbitrary_data blob
) engine=your_engine;

and let’s imagine that we need to process this query:

INSERT INTO t1 VALUES (1, "hello world", "blobbbbb");

Now, the storage engine needs to “pack” the data for each column into it’s buffer in the write_row() function. Conveniently, Drizzle/MySQL provides a pack() function for it’s column types (fields) that will do the data packing for you. That is, you do not have to inspect the provided row for pointers to the actual data and do the packing/copying yourself.

How? well, the table object (which is visible from your engine) conveniently holds a list of fields in the appropriate order. The actual pack() function is a member of these fields so you just need to call it as you loop over the list:

/* make sure row_buffer has enough memory */
unsigned char *pos = row_buffer;
 
/* copy NULL bits, "table->s" is the TableShare object */
memcpy(pos, row, table->s->null_bytes);
pos += table->s->null_bytes;
 
/* "row" is the MySQL formatted row given by the core */
for (Field **field = table->field; *field; field++) {
  if (!((*field)->is_null()))
    pos = (*field)->pack(pos, row + (*field)->offset(row));
}

The above code snippet will populate “row_buffer” with the actual data that you want to write to your storage mechanism. You do not have to forward the “pos” pointer because pack() returns a pointer at the end of where it had worked in the buffer (think Pascal Strings). This is precisely why we created the pos pointer, to avoid row_buffer from being forwarded.

For the opposite situation (when retrieving a row), an unpack() function is provided for each field so you just need to take advantage of it like we did with the pack() snippet above.

Little bit more on fields

The actual pack() function that gets called depends on the type of column since the Field class is an abstract base class for the sub classes that actually represents column types inside Drizzle/MySQL. If you want to know what a pack() function looks like for a BLOB type, grep for “Field_blob” in the source tree and there will be a pack() member function for it.

The code layout for field subsystem in MySQL is rather difficult to comprehend since everything is crammed in “sql/field.c” and “sql/field.h” files (at least as of 5.4). So, if you want to get a good grasp of how things are architectured, you should take a look at Drizzle. Field subclasses are located individually in the “drizzled/field/” directory and the base class is located in “drizzled/field.h”.

So, that’s about it! Hopefully this information will help other engine developers when they come across a need to support variable width tables :)

Toru Maesaka drizzle, knowledge, oss drizzle, engine, mysql, storage

If you’re working on developing a Drizzle plugin, you may come across situations where you want to accept user options for it at server startup. For example, if you design your plugin to create files for activity logging, you may want to allow the DBA to specify where to write those files out.

In my case, I decided to provide a command line option to BlitzDB for row based query caching. This option is intended for special use-cases where the read/write ratio is 9:1. For those that are interested, row caching is disabled by default because it creates overhead in the engine for read-through logic and cache invalidation _unless_ read requests are significantly higher than update requests.

There are situations where BlitzDB’s row cache can be helpful but this is beyond the scope of this entry so I will save it for another day :)

Adding startup options to your plugin

Drizzle allows you to add command line options to your plugin without editing the server code. But before you start hacking away, there are few not-so-obvious things that you need to understand.

So, let us first look at the data types that your plugin can accept:

• DRIZZLE_SYSVAR_BOOL
• DRIZZLE_SYSVAR_STR
• DRIZZLE_SYSVAR_INT
• DRIZZLE_SYSVAR_UINT
• DRIZZLE_SYSVAR_LONG
• DRIZZLE_SYSVAR_ULONG
• DRIZZLE_SYSVAR_LONGLONG
• DRIZZLE_SYSVAR_ULONGLONG
• DRIZZLE_SYSVAR_ENUM

As you can see, there is a wide range of types that you can choose from. What you should choose depends on what you want to use the value for.

Pick your data type

So lets take my row cache option as an example. Caching over 4 billion rows in one physical server is very unlikely and since we’re not interested in negative numbers, we’re going to pick:

• DRIZZLE_SYSVAR_UINT

which we can store the value as uint32_t in the plugin.

Declare that your plugin accepts options

Every plugin must declare itself as a plugin which looks like this for BlitzDB:

drizzle_declare_plugin(blitz) {
  "BLITZ",
  "0.3",
  "Toru Maesaka",
  "Non-transactional General Purpose Engine",
  PLUGIN_LICENSE_GPL,
  blitz_init,             /*  Plugin Init      */
  blitz_deinit,           /*  Plugin Deinit    */
  NULL,                   /*  status variables */
  blitz_system_variables, /*  system variables */
  NULL                    /*  config options   */
}
drizzle_declare_plugin_end;

Here, we’re interested in the second last argument which is called blitz_system_variables in the above example. Feel free to call this whatever you like for your plugin.

So what exactly is blitz_system_variables? Its a null-terminated array of system variables that your plugin accepts. This is what it looks like for BlitzDB:

static struct st_mysql_sys_var *blitz_system_variables[] = { 
  DRIZZLE_SYSVAR(row_cache),
  NULL
};

As you can see, BlitzDB only supports one option at the moment so there is only one entry called row_cache.

Define your options

You must define every option that you’ve added to the system variable array. We decided to use DRIZZLE_SYSVAR_UINT earlier and called it row_cache so it is defined like this:

static DRIZZLE_SYSVAR_UINT (
  row_cache, /* option name */
  blitz_row_cache_size, /* variable to set the value to */
  PLUGIN_VAR_READONLY, /* mode */
  N_("Enable row caching for BlitzDB tables."),
  NULL,       /*  check func    */
  NULL,       /*  update func   */
  0,          /*  default value */
  0,          /*  minimum value */
  UINT32_MAX, /*  maximum value */
  0           /*  block size    */
);

The comments pretty much explains what the arguments are but for more details, you should take a look at the macros in drizzled/plugin.h. You could also look at what other plugins do by grepping for the system variable type that you’re interested in.

Test your new startup option

If all goes well you should be able to compile Drizzle and check whether command line options are visible from the plugin. An option takes the following form:

--<name_of_plugin>-<option_name>

So, in the row cache example, row cache can be enabled like this:

/usr/local/sbin/drizzled --blitz-row_cache=10000

Also note that you can replace the underscore with a hyphen:

/usr/local/sbin/drizzled --blitz-row-cache=10000

That’s it! it should be relatively easy to add more options once you successfully get your first one done.

Toru Maesaka drizzle, knowledge, oss blitzdb, drizzle

Tokyo Cabinet (TC) provides iteration functionality for both it’s persistent and non-persistent data structures. For example, if you wanted to iterate through TC’s hash database, you can use the tchdbiternext() function. This is really straight forward to use such that:

void *key;
int key_len;
 
if (tchdbiterinit(tc_database_handle) != true) {
  /* failed to initialize iterator */
}
 
while ((key = tchdbiternext(tc_database_handle, &key_len)) != NULL) {
  /* work with the fetched key and key_len */
}

will iterate through the entire hash database that “tc_database_handle” object is responsible for. This can be handy if you need to loop through your database for some arbitrary reason.

However, there is a consequence in using this function in a concurrent environment with a use-case where the order of records _really_ matter. This is because even though TC is a thread-safe library, the iteration functions aren’t thread-safe in a way that we expect.

For example, if a write operation occurs while the application iterates over the database, you will end up iterating over a database that is in a changed state. This will not make the cursor go crazy and crash your application since TC handles this internally but you still end up iterating over a database that is in a state that you did not initially intend on looping through.

Solution to this is to simply block write operations to the database while your application iterates through. For example, you could use pthread’s rw_lock to allow other threads to read while you iterate but block writes until you finish iterating.

I was planning on doing this for a table scanner in the storage engine that I’m currently working on but turns out TC has an undocumented function that will take care of this internally. I’ve talked to Mikio about this function and apparently it is intentional that he hasn’t documented it on his specification page. He has no plans on throwing it out so you do not have to worry about it to magically disappear one day. For more information, you can take a look at his header file (tchdb.h for hash database).

Explanation and Simple Example

The function is called tchdbforeach() which will atomically iterate through your database from beginning to the end by supplying each key/value pair to the callback function that you provide. The signature of the callback is the following:

bool callback(const void *kbuf, int ksiz, const void *vbuf,
              int vsiz, void *op);

where the fifth argument, “void *op” is an opaque pointer to the data that you can pass to the callback. Here is a simple example that will increment a counter integer on each iteration using this function:

/* Do whatever you like with the provided key/value pair in here */
bool callback(const void *kbuf, int ksiz, const void *vbuf,
              int vsiz, void *op) {
  if (op == NULL)
    return false;
 
  *((int *)op) += 1;
 
  return true;
}
 
int main(void) {
  int niter = 0;
 
  ...
 
  if (!tchdbforeach(tc_database_handle, callback, &niter)) {
    fprintf(stderr, "failed to iterate the database\n");
    return EXIT_FAILURE;
  }
 
  printf("iterated %d times\n", niter);
 
  ...
 
  return EXIT_SUCCESS:
}

If all goes well, the counter variable will be set to the number of records in the database. This function is slightly more complex than using tchdbiternext() but you are guaranteed to iterate atomically which is pretty important for a table scanner.

I hope this function can help you too.

Toru Maesaka knowledge, oss storage, tip, tokyocabinet

I’ve decided to start a series of blog entries on not-so-obvious findings that I’ve found while working on my new project. By archiving the findings, I’m hoping that I can help those that are looking into developing a storage engine for the MySQL family in the future.

Accumulating these mini-knowledge would also be useful for me since I can refer back to it when I forget something. Also, once I write enough entries I’m planning on summarizing them and making it available on the Drizzle Wiki. If MySQL is interested in updating the engine documentation, I would be more than happy to help there too.

So to begin with, I’ll describe something trivial that I stumbled across while trying to catch an error on duplicate primary key insertion to the data table.

Background

In brief, the database kernel does not care if the INSERT query contains a duplicate primary key for a given table or not. It is the storage engine’s job to tell the kernel that the request was invalid due to key collision. If a storage engine fails to do this, the kernel will acknowledge that the query was successful (given that no other errors were thrown) and will keep doing what it needs to do.

Mechanics

Data insertion is handled inside the write_row() function that your engine must implement. The return value of this function is an integer that represents the status of the work it had done. After looking through the possible error statuses in “drizzled/base.h”, I immediately found this:

#define HA_ERR_FOUND_DUPP_KEY 121 /* Dupplicate key on write */

I also looked through MyISAM and InnoDB to confirm that this was indeed the correct error status to return on duplicate primary key. Here is the snippet of my row insertion at the time:

/* TC's tchdbputkeep will not insert a row to the table if there
   was a collision */
if (tchdbputkeep(data_table, primary_key, primary_key_length, buf,
                 table->s->reclength) == false) {
  my_errno = HA_ERR_GENERIC;
 
  /* check for primary key collision */
  if (tchdbecode(data_table) == TCEKEEP)
    my_errno = HA_ERR_FOUND_DUPP_KEY;
 
  return my_errno;
}

On first glimpse, this seems right but the error I was getting from the command line prompt always differed with MyISAM and InnoDB despite returning the same error status. Specifically, this is what I was getting:

ERROR 1022 (23000): Can't write; duplicate key in table 't1'

whereas I was getting this error on other engines:

ERROR 1062 (23000): Duplicate entry '1' for key 'PRIMARY'

At this stage I couldn’t make sense of what I was doing wrong but it turned out that the solution was pretty simple.

Solution

After talking to Stewart Smith about my issue in #drizzle @ freenode, it turned out I am supposed to keep track of which key the duplication was found in write_row() and inform it to the kernel via the info() function.

You can do this by setting the errkey integer variable to the key number that is used internally by the kernel. So, obtaining the internal primary key number with this call in write_row():

share->errkey = table->s->primary_key;

and adding the following code to info():

if (flag & HA_STATUS_ERRKEY) {
  errkey = share->errkey;
}

happily fixed the issue I was experiencing. Yay.

I guess reading the section on info() in the document gives a hint that this is where you supply the key number on key-error but frankly, this is really easy to forget and miss since the importance isn’t so emphasized.

Anyhow, thats all I have to say in the first of this series and hopefully I’ll write something more interesting in the upcoming entries. Until then, happy hacking ;)

Toru Maesaka drizzle, knowledge, oss drizzle, engine, mysql, storage