Major system upgrade scheduled
Posted: Sat Feb 08, 2020 12:35 pm
A major system upgrade has been scheduled within the next seven weeks. We anticipate the upgrade to occur no later than April 10, but is most likely to occur during the last week of March. We'll post an update about the exact date when it is available. The downtime is expected to last between 24 and 48 hours. We cannot justify raising fees to purchase a backup system for rare instances like this, so all services will be offline.
Some of the upgrades are tasks that we normally perform yearly, but did not do during our normal pre-Thanksgiving period last year. More importantly, we need to make some major changes that we have not done since the system was brought online in December 2013.
First, we will be relocating the servers to a new datacenter closer to where we work. The company where the servers are currently located is experiencing financial difficulties, and we don't want to be surprised with a sudden bankruptcy. Unlike many cloud-based pools, we own and operate our own physical servers, so that there is zero risk that the cloud provider's employees can be unethical or can be tricked into granting access. We will shut down the entire server rack, drive a truck to the new location, and reconfigure it for new IP addresses there.
Second, we will be doubling both our downstream and upstream bandwidth so that coins will be more likely to stay connected to their networks and have lower orphan rates.
Third, we will be upgrading the system to Postgres 12, the latest version of the database software.
Fourth, Chris will be changing some solid state disks to use the TRIM command, which was incorrectly set up when the drives were originally formatted, and which causes the disks to slow down over time. We need to copy all the data to hard drives, reformat the disks, and then copy it back.
Fifth, Chris will upgrade some servers from Debian 6 and 7, which are obsolete, to the latest version of the operating system.
Finally, there are some database bugs that need to be resolved by huge queries that would lock up the system if people were connected to it. For example, we discovered that the "worker history" charts on the "details" page incorrectly record worker difficulties for some algorithms, because the column type is "integer" instead of "double." The table has billions of rows and the query to update them requires an hour.
We'll update everyone as the date approaches for the upgrades. About a month after the upgrades are finished, we anticipate deploying a brand new, easier-to-use website with a modern look and feel.
Some of the upgrades are tasks that we normally perform yearly, but did not do during our normal pre-Thanksgiving period last year. More importantly, we need to make some major changes that we have not done since the system was brought online in December 2013.
First, we will be relocating the servers to a new datacenter closer to where we work. The company where the servers are currently located is experiencing financial difficulties, and we don't want to be surprised with a sudden bankruptcy. Unlike many cloud-based pools, we own and operate our own physical servers, so that there is zero risk that the cloud provider's employees can be unethical or can be tricked into granting access. We will shut down the entire server rack, drive a truck to the new location, and reconfigure it for new IP addresses there.
Second, we will be doubling both our downstream and upstream bandwidth so that coins will be more likely to stay connected to their networks and have lower orphan rates.
Third, we will be upgrading the system to Postgres 12, the latest version of the database software.
Fourth, Chris will be changing some solid state disks to use the TRIM command, which was incorrectly set up when the drives were originally formatted, and which causes the disks to slow down over time. We need to copy all the data to hard drives, reformat the disks, and then copy it back.
Fifth, Chris will upgrade some servers from Debian 6 and 7, which are obsolete, to the latest version of the operating system.
Finally, there are some database bugs that need to be resolved by huge queries that would lock up the system if people were connected to it. For example, we discovered that the "worker history" charts on the "details" page incorrectly record worker difficulties for some algorithms, because the column type is "integer" instead of "double." The table has billions of rows and the query to update them requires an hour.
We'll update everyone as the date approaches for the upgrades. About a month after the upgrades are finished, we anticipate deploying a brand new, easier-to-use website with a modern look and feel.