Maintenance on Tuesday, September 12

[Steve Sokolowski]

One of our goals is to, every weekend, make one or two performance improvements. Even though the system is running great now, the goal is to always stay ahead of the possibility of an increasing number of customers causing a system crash at some indefinite point in the future.

One of the ways we have been improving performance is by changing the types of some columns into smaller datatypes. On Tuesday, we plan to continue this practice by changing coin_id from 8 bytes to 2 bytes in the shares tables. With these tables containing hundreds of millions of rows, a seemingly minor change like this can reduce the size of the database by 4GB and reduce the amount of data that needs to be written to the disks when shares are inserted.

To accomplish this change, we need to lock the table and rewrite it. The process will take about two hours and will occur sometime on Tuesday.

• Mining will not be affected.
• Some charts on the website will fail to load during the maintenance period.
• Earnings will appear to fall behind during the maintenance period.
• When the maintenance period is over, any data that was delayed in being inserted into the database will be inserted over the course of 15 minutes.

If you aren't looking at the website during this maintenance period, you may not notice anything at all. Regardless, no action by miners is necessary other than recognizing that balances and charts will be out of date for a few hours.

[Steve Sokolowski]

The database seems to be steady at about 48 minutes behind right now, so the latest stats you're seeing are what happened 48 minutes ago. The data since then is queued in memory and is being inserted as quickly as possible. It will fully catch up at some point later today.

Also, there will be a brief mining server restart tonight to add some code that will prevent against pool losses by checking exchange block numbers and erroring out coins that cannot be sold.

[claydigger]

Steve,
Thanks for the transparency of the activities, the detailed explanations as to the nature of the changes and for pushing yourselves to futureproof the site for new growth.
claydigger

[Steve Sokolowski]

The server restart was delayed because we discovered that our script that deletes old shares is now performing too slowly for the increased hashrate recently. It never finished in time for the restart because it took all day to delete the shares.

That caused the database to be behind all day yesterday, and statistics were continually out of date. Fortunately, no money was lost for anyone and we figured out how to delete shares more quickly next week to prevent this from happening again. We'll need to perform the restart tonight to get the more accurate trading code out.