Thu, Jun 5, 2008 02:38 AM
So, we had a cron job hanging for hours. No idea why. So, I started debugging. It all came down to a call to in_array(). See, this job is importing data from a huge XML file into MySQL. After it is done, we want to compare the data we just added/updated to the data in the table so we can deactivate any data we did not update. We were using a mod_time field in mysql in the past. But, that proved to be an issue when we wanted to start skipping rows from the XML that were present but unchanged. Doing that saved a lot of MySQL writes and sped up the process.
So, anyhow, we have this huge array of ids accumulated during the import. So, an in clause with 2 million parts would suck. So, we suck back all the ids in the database that exist and stick that into an array. We then compared the two arrays by looping one array and using in_array() to check if the value was in the second array. Here is a pseudo example that shows the idea:
[sourcecode language='php']
foreach($arr1 as $key=>$i){
if(in_array($i, $arr2)){
unset($arr1[$key]);
}
}
[/sourcecode]
So, that was running for hours with about 400k items. Our data did not contain the value as the key, but it could as the value was unique. So, I added it. So, now, the code looks like:
[sourcecode language='php']
foreach($arr1 as $key=>$i){
if(isset($arr2[$i])){
unset($arr1[$key]);
}
}
[/sourcecode]
Yeah, that runs in .8 seconds. Much better.
So, why were we using in_array to start with if in_array is clearly not the right solution to this problem? Well, it was basic code evolution. Originally, these imports would be maybe 100 items. But, things changed.
FWIW, I tried array_diff() as well. It took 25 seconds. Way better than looping and calling in_array, but still not as quick as a simple isset check. There was refactoring needed to put the values into the keys of the array.
UPDATE: I updated this post to properly reflect that there is nothing wrong with in_array, but simply that it was not the right solution to this problem. I wrote this late and did not properly express this. Thanks to all those people in the comments that helped explain this.
Mon, Jun 2, 2008 09:59 PM
SimpleXML is neat. Some people don't think it is so simple. Boy, use the old stuff. The DOM-XML stuff.
Anyhow, one annoying thing about SimpleXML has to do with caching. When using web services, we often cache the contents we get back. We were having a problem where we would get an error about a SimpleXML node not existing. We were caching the data in memcached which serializes the variable. So, when it unserialized the variable, there were references in there to some SimpleXML nodes that we did not take care of. Basically, a tag like:
<foo>bar</foo>
is a string. But a tag like:
<foo></foo>
is an empty SimpleXML Object. That is a little annoying, but I don't feel like digging into the C code and figuring out why. So, we just work around it. We made a recursive function to do the dirty work for us.
function makeArray($obj) {
$arr = (array)$obj;
if(empty($arr)){
$arr = "";
} else {
foreach($arr as $key=>$value){
if(!is_scalar($value)){
$arr[$key] = makeArray($value);
}
}
}
return $arr;
}
That will turn whatever you pass it into an array or empty string if it is empty.
But, while I was hacking around tonight, I came up with another idea. Check out this hackery:
$data = json_decode(json_encode($data));
Yeah! One liner. That converts all the SimpleXML elements into stdClass objects. All other vars are left intact.
Ok, so this is where someone in the comments can tell me about the magic SimpleXML method or magic OOP function I have missed to take care of all this. Go ahead, please make my code faster. I dare you.
Wed, May 28, 2008 09:18 AM
So, I was asked in IRC today about the proposed short array syntax for PHP. For those that don't know, I mean the same syntax that other languages (javascript, perl, python, ruby) all have. Currently in PHP we have this:
$var = array(1,2,3);
The proposed additional syntax is:
$var = [1,2,3];
So, I voted +1 for this feature on the PHP Internals list. A colleague asked me why I voted +1. At first I had no good answer other than it was just a gut feeling. It just feels like a good addition to the language. It is common among web languages and therefore users coming into PHP from other languages may find it more comfortable.
The best thing I could tell him was that it would make arrays fall in line with other data types in PHP. For example, you never write:
$var = int(1);
$var = string(foo);
So, why oh why do we have to have what looks like a function, but in reality is not, for creating an array? It is a language construct and should look like a language construct. I think the [ ] syntax makes more sense when you think about it in those terms.
I say commit it Andi. That seems to be what everyone else does. =)
Wed, May 7, 2008 12:48 PM
Well, it has been almost a month. I know I am late to the blogosphere on my thoughts. Just been busy.
Again this year, the Phorum team was invited to be a part of the DotOrg Pavilion. What is that? Basically they just give expo floor space to open source projects. It is cool. We had a great location this year. We were right next to the area where they served food and drinks during the breaks. We had lots of traffic and met some of our power users. IMVU.com is getting 1.5 million messages per month in their Phorum install. They did have to customize it to fit into their sharding. But, that is expected. A guy (didn't catch his name) from Innobase came by and told us that they just launced InnoDB support forums on their site using Phorum. Cool. So now MySQL and Innobase use Phorum. I am humbled by the message that sends to me about Phorum.
Speaking of our booth, we were right next to the phpMyAdmin guys. Wow, that product has come a long way. I was checking out the visual database designer they have now. It was neat. I also met the Gentoo MySQL package maintainer. He was in the phpMyAdmin booth.
I was interviewed by WebDevRadio as I already posted. I was also asked to do a short Q&A with the Sun Headlines video team. They used one part of my clip. I won't link to that. No, if you find it good for you. I need to be interviewed some more or something. I did not look comfortable at all.
There were lots of companies with open in their name or slogan. I guess this is expected pandering.
I attended part of the InnoDB talk given by Mark Callaghan of Google. It appears that Google is serious about improving InnoDB on large machines. That is, IMO, good news for anyone that likes InnoDB. If I counted right, they had more than 5 people who at least part of their job is to improve InnoDB.
I gave my two talks. The first had low attendance, but the feedback was nice. It was just after the snack break in the expo hall and I was in the farthest room from the expo hall. That is what I keep telling myself. =) The second was better attended and the feedback seemed good there. I was told by Maurice (Phorum Developer) that I talked too fast and at times sounded like Mr. Mackey from South Park by repeating the word bad a lot. I will have to work on that in the future. I want to do more speaking.
On the topic of my second talk, there seemed to be a lot of "This is how we scaled our site" talks. I for one found them all interesting. Everyone solves the problem differently.
Next year I am thinking about getting more specific with my talk submissions. Some ideas include: PHP, MySQL and Large Data Sets, When is it ok to denormalize your data?, Using memcached (not so much about how it works), Index Creation (tools, tips, etc.).
In closing, I want to give a big thanks to Jay Pipes and Lenz Grimmer from MySQL. Despite Jay's luggage being lost he was still a big help with some registration issues among other things. Both of them helped out the Phorum team a great deal this year. Thanks guys.
Tue, May 6, 2008 11:03 AM
A coworker found out how secure Amazon's MP3 store is. Even big guys like Amazon make errors in their web site security.
So, I clicked purchase and the album immediately started downloading. It was at this point that I had the thought cross my mind: "Did I update my credit card info?"
Well, no, I didn't. Before the album finished downloading, I was trying to change the method of payment. Turns out, for a digital purchase, you can't do such a thing. So, I waited and wondered was was going to come of this...
Tue, May 6, 2008 10:17 AM
NEW UPDATE: MySQL Forge is being end of lifed. And honestly, there were never that many examples there. It just never took off. But, there is good news. Percona, the company in my opinion that is leading the way for binary compatible MySQL progress, has an online tool for creating a my.cnf file based on your input about your server, your data and your workload. It seems to work very well. I recommend using it for creating your my.cnf file.
UPDATE: There are some examples being added at the MySQL Forge now.
When I first started installing MySQL for myself, it was quite handy to have the example my.cnf files in the source package. I was a noob to the MySQL configuration. Even after I became more experienced, I would use them as a starting point. However, I now find that they are so behind the times they are not as useful. Here are some of the comments from the files.
my-small.cnf
# This is for a system with little memory (<= 64M) where MySQL is only used
# from time to time and it's important that the mysqld daemon
# doesn't use much resources.
my-medium.cnf
# This is for a system with little memory (32M - 64M) where MySQL plays
# an important part, or systems up to 128M where MySQL is used together with
# other programs (such as a web server)
my-large.cnf
# This is for a large system with memory = 512M where the system runs mainly
# MySQL.
my-huge.cnf
# This is for a large system with memory of 1G-2G where the system runs mainly
# MySQL.
I end up using the large or huge files as a starting point for every server I set up by hand. The small and medium should be renamed underpowered and teeny-tiny. Who has less than 64MB of RAM on a server now? Can you even buy sticks of memory that small in any modern system? Most come with 256MB sticks minimum. And they never come with just one stick.
I will use the large example as a starting point for a server that has 2GB of RAM and will be running an entire site on one server. I use huge for any server that runs only MySQL. And even then, most of them have 4GB of RAM or more.
I don't know if anyone at MySQL has plans on tweaking these files or not. Perhaps those good guys at the MySQL Performance Blog or Percona could create some example my.cnf files. I could put some out there, but I fear their sole purpose would be for someone to point out what I am doing wrong. =P Hey, they work for me. Hmm, maybe this would make a good MySQL Forge section. A whole area of user contributed my.cnf files. They could be architecture specific and everything. What runs best on Solaris? Linux? BSD? Windows? 32-bit? 64-bit?
One thing I would for sure like to see is example files for InnoDB dominant servers. Most of our servers all run primariy InnoDB tables. None of these above examples covers InnoDB. They have comments, but no preconfigured values. I have seen more than one server using InnoDB tables without any custom configuration in their my.cnf. In the end that is the fault of the server admin/owner no doubt.
What do you say? Anyone up for a MySQL Forge section for my.cnf files?
Sat, May 3, 2008 10:13 AM
While I was at the MySQL Conference, I sat down with Michael Kimsal of WebDevRadio and recapped the two talks that I gave at the conference. I have uploaded the slides so you can follow along if you want.
One to a Cluster - The evolution of the dealnews.com architecture.
MySQL Tips and Tricks - Some simple tips and some of the more advanced SQL we use in Phorum.
Thanks Michael. Any time you need a guest, just let me know.
Thu, May 1, 2008 01:09 PM
I was reading a post about The Swag Report and realized that I stayed so busy at the Phorum booth (and a little at the memcached booth) and preparing for my talks, I did not bother to go around and collect any swag from the conference. So, if you are a vendor and want to mail me some swag that I missed, you can send it to: Brian Moon, 198 S. Hillcrest Rd., Odenville, AL 35120. Of course, I expect nothing. But, ya never know what product I might pimp because of a t-shirt. =)
Tue, Apr 22, 2008 01:44 PM
So, I mentioned before that I found out about index_merge at the MySQL Conference. I was wondering why I had not heard more about it since it came out in 5.0.3. When talking with some MySQL people about it, I received mixed results. So, I decided to kind of run my own tests on some data and see what I could figure out.
I apologize for Wordpress' bad output. =(
The Data
I created a table with 5 million rows. Early tests with MySQL's Harrison Fisk (HarrisonF) over my shoulder with small data sets showed MySQL would optimize out the indexes in favor of table scans. I wanted to avoid that. This is my table schema:
CREATE TABLE `test2` (
`id1` int(10) unsigned NOT NULL default '0',
`id2` int(10) unsigned NOT NULL default '0',
`id3` int(10) unsigned NOT NULL default '0',
`dt` datetime NOT NULL default '0000-00-00 00:00:00',
`somevar` varchar(255) NOT NULL default '',
KEY `id1` (`id1`),
KEY `id2` (`id2`)
) ENGINE=MyISAM
The field id1 was filled with random vaules between 1 and 5000. I filled id2 with random values between 1 and 100, except that about half the data has the value 999 in it. This was to emulate the issue we were seeing on the smaller table. We found that if a value was in more than n% of the rows, the optimizer would skip the index. I wanted to test that on larger data sets. id3 was filled with random values between 1 and 1000000. dt was a random date/time between 1999 and 2008. and somevar was a random string chars.
Intersect Merges
mysql> explain select count(*) from test2 where id2=99 and id1=4795;
+----+-------------+-------+-------------+---------------+---------+---------+------+------+----------------------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+-------------+---------------+---------+---------+------+------+----------------------------------------------------+
| 1 | SIMPLE | test2 | index_merge | id1,id2 | id1,id2 | 4,4 | NULL | 3 | Using intersect(id1,id2); Using where; Using index |
+----+-------------+-------+-------------+---------------+---------+---------+------+------+----------------------------------------------------+
This is the most basic of example. MySQL uses the two indexes, finds where they intersect and merges the data together. This query is quite fast, although a key on the two together would be faster. If you have this showing up a lot, you probably need to combine the two keys into one. I should also note that in this example, only the keys are needed, no data from the tables. This is important.
mysql> explain select sql_no_cache somevar from test2 where id2=99 and id1=4795;
+----+-------------+-------+------+---------------+------+---------+-------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+------+---------------+------+---------+-------+------+-------------+
| 1 | SIMPLE | test2 | ref | id1,id2 | id1 | 4 | const | 930 | Using where |
+----+-------------+-------+------+---------------+------+---------+-------+------+-------------+
As you see, as soon as we ask for data that is not in the indexes, our intersect is dropped in favor of using the key with the least values and simply scanning on those to match the rest of the where clause. This was the case pretty much every time I tried it. I was never able to use an index_merge with intersect when requesting data not available in the key.
Union Merges
explain select sql_no_cache somevar from test2 where id2=99 or id1=4795;
+----+-------------+-------+-------------+---------------+---------+---------+------+-------+-----------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+-------------+---------------+---------+---------+------+-------+-----------------------------------+
| 1 | SIMPLE | test2 | index_merge | id1,id2 | id2,id1 | 4,4 | NULL | 27219 | Using union(id2,id1); Using where |
+----+-------------+-------+-------------+---------------+---------+---------+------+-------+-----------------------------------+
mysql> select sql_no_cache somevar from test2 where id2=99 or id1=4795;
26237 rows in set (0.20 sec)
This merge type takes to keys involved in an OR and then merges the data much like a UNION statement would. As you can see, in this case, it did use the index even though we requested `somevar` that is not in the index.
To show the alternative to this, I selected using id3 instead of id1. id3 has no index.
mysql> explain select sql_no_cache somevar from test2 where id2=99 or id3=266591;
+----+-------------+-------+------+---------------+------+---------+------+---------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+------+---------------+------+---------+------+---------+-------------+
| 1 | SIMPLE | test2 | ALL | id2 | NULL | NULL | NULL | 5000000 | Using where |
+----+-------------+-------+------+---------------+------+---------+------+---------+-------------+
mysql> select sql_no_cache somevar from test2 where id2=99 or id3=266591;
25252 rows in set (26.01 sec)
As you can see, this does a table scan even though there is a key on id2. It does you know good.
Sort Union Merge
mysql> explain select sql_no_cache id1, id2 from test2 where id2=99 or id1 between 4999 and 5000;
+----+-------------+-------+-------------+---------------+---------+---------+------+-------+----------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+-------------+---------------+---------+---------+------+-------+----------------------------------------+
| 1 | SIMPLE | test2 | index_merge | id1,id2 | id2,id1 | 4,4 | NULL | 44571 | Using sort_union(id2,id1); Using where |
+----+-------------+-------+-------------+---------------+---------+---------+------+-------+----------------------------------------+
mysql> select sql_no_cache somevar from test2 where id2=99 or id1 between 4999 and 5000;
27295 rows in set (0.19 sec)
This behaves much like the union merge. However, because one index is using a range, MySQL must first sort one index and then merge the two. Again, if I switch this to an AND instead of an OR, index_merge is not used in favor of scanning the id2 indexed data for matches to the rest of the where clause.
Conclusion
Hmm, after all this, I see why this was not a big announcement. It can only make bad SQL and tables better. Tables and queries that are already optimized using composite indexes will see no benefit from this. At best this will help me with some one off queries or reports that are only run monthly where I don't want to pollute the indexes with special cases just for those queries.
Thu, Apr 17, 2008 04:43 PM
It is always surprising what I learn when I go to a conference these days. Years ago, I could go to any talk and just suck it all in. Now, it is the little nuggets. The topics as a whole do more to confirm what I have already developed while running the Phorum project and building the infastructure for dealnews.com. That confirmation is still nice. You know you are not the only one that thought a particular solution was a good idea.
One of the confirmations I have had is that the big sites like Flickr, Wikipedia, Facebook and others don't use exotic setups when it comes to their hardware and OS. During a keynote panel, they all commented that they did not do any virtualization on their servers. Most did not use SANs. Some ran older MySQL versions but some were running quite recent versions. I have kept thinking that I did not have the desire to get to fancy with that stuff and clearly I am not the only one.
One of the little nuggets that will likely change my world is index_merge in MySQL. I feel silly as this has been around since 5.0.3 but I was not aware of it. Basically MySQL will now use more than one key to resolve a where clause and possibly an order by depending on the query. This could lead to me removing several keys from tables in both Phorum and at dealnews.
There were others, but I am tired and trying to get OpenID into the Phorum trunk right now so I will have to think of more later.
