The Question Recently, a customer asked us:

Why would Tungsten Replicator be slow to apply to MySQL?

The Answer Performance Tuning 101

When you run trepctl status and see:
appliedLatency : 7332.394
like this on a slave, it is almost always due to the inability for the target database to keep up with the applier.

This means that we often need to look first to the database layer for the solution.

Here are some of the things to think about when dealing with this issue:

Architecture and Environment
√ Are you on bare metal?
√ Using the cloud?
√ Dev or Prod?
√ Network speed and latency?
√ Distance the data needs to travel?
√ Network round trip times? Is the …

Overview The Skinny

From time to time we are asked how to check whether or not there are data discrepancies between Master/Slave nodes within a MySQL (or MariaDB) cluster that’s managed with Tungsten Clustering. This is always a challenging task, not least because we hope and believe that our replication mechanism would avoid such occurrences, that said there can be factors outside of our control that can appear to “corrupt” data – such as inadvertent execution of DML against a slave using a root level user account.

Tungsten Replicator, the core replication component in our Tungsten Clustering solution for MySQL (& MariaDB), is just that, a replicator – it takes transactions from the binary logs and replicates them around. The replicator isn’t a data synchronisation tool in that respect, the …

The Question Recently, a customer asked us:

If we move to using the INI configuration method instead of staging, would password-less SSH still be required?

The Answer The answer is both “Yes” and “No”

No, for installation and updates/upgrades specifically. Since INI-based configurations force the tpm command to act upon the local host only for installs and updates/upgrades, password-less SSH is not required.

Yes, because there are certain commands that do rely upon password-less SSH to function. These are:

• tungsten_provision_slave
• prov-sl.sh
• multi_trepctl
• tpm diag (pre-6.0.5)
• tpm diag --hosts (>= 6.0.5)
• Any tpm-based backup and restore operations that involve a remote node

Summary The Wrap-Up

In …

Overview The Skinny

In this blog post we will discuss how to best integrate various Continuent-bundled cluster monitoring solutions with PagerDuty (pagerduty.com), a popular alerting service.

Agenda What’s Here?

• Briefly explore the bundled cluster monitoring tools
• Describe the procedure for establishing alerting via PagerDuty
• Examine some of the multiple monitoring tools included with the Continuent Tungsten Clustering software, and provide examples of how to send an email to PagerDuty from each of the tools.

Exploring the Bundled Cluster Monitoring Tools A Brief Summary

Continuent provides multiple methods out of the box to monitor the cluster health. The most popular is the suite of Nagios/NRPE scripts (i.e. cluster-home/bin/check_tungsten_*). We also have Zabbix scripts (i.e. cluster-home/bin/zabbix_tungsten_*). Additionally, there is …

Cluster management, a common software infrastructure among technology companies, aggregates compute resources from a collection of physical hosts into a shared resource pool, amplifying compute power and allowing for the flexible use of data center hardware. At Uber, cluster management …

The post Peloton: Uber’s Unified Resource Scheduler for Diverse Cluster Workloads appeared first on Uber Engineering Blog.

One important way to protect your data is to keep your Tungsten Clustering software up-to-date.

A standard cluster deployment uses three nodes, which allows for no-downtime upgrades along with the ability to have a fully available cluster during maintenance.

Please note that with only two database cluster nodes, there is a window of vulnerability created by leaving zero failover candidates available when the lone slave is taken down for service.

The Best Practices: Staging Performing a No-Downtime Upgrade for a Staging Deployment

When upgrading a Staging-style deployment, all nodes are upgraded at once in parallel via the tools/tpm update command run from inside the staging directory on the staging host.

No Master switch happens, and all layers are restarted to use the new code. …

Tungsten Clustering is an extraordinarily flexible tool, with options at every layer of operation.

In this blog post, we will describe and discuss the two different methods for installing, updating and upgrading Tungsten Clustering software.

When first designing a deployment, the question of installation methodology is answered by inspecting the environment and reviewing the customer’s specific needs.

Staging Deployment Methodology

All for One and One for All

Staging deployments were the original method of installing Tungsten Clustering, and relied upon command-line tools to configure and install all cluster nodes at once from a central location called the staging server.

This staging server (which could be one of the cluster nodes) requires SSH access to all …

Uber is committed to delivering safer and more reliable transportation across our global markets. To accomplish this, Uber relies heavily on making data-driven decisions at every level, from forecasting rider demand during high traffic events to identifying and addressing bottlenecks…

The post Uber’s Big Data Platform: 100+ Petabytes with Minute Latency appeared first on Uber Engineering Blog.

Your database cluster contains your most business-critical data and therefore proper performance under load is critical to business health. If response time is slow, customers (and staff) get frustrated and the business suffers a slow-down.

If the database layer is unable to keep up with demand, all applications can and will suffer slow performance as a result.

To prevent this situation, use load tests to determine the throughput as objectively as possible.

In the sample load.pl script below, increase load by increasing the thread quantity.

You could also run this on a database with data in it without polluting the existing data since new test databases are created to match each node’s hostname for uniqueness.

Note: The examples in this blog post assume that a Connector is …

In a previous post we went into detail about how to implement Tungsten-specific checks. In this post we will focus on the other standard Nagios checks that would help keep your cluster nodes healthy.

Your database cluster contains your most business-critical data. The slave nodes must be online, healthy and in sync with the master in order to be viable failover candidates.

This means keeping a close watch on the health of the databases nodes from many perspectives, from ensuring sufficient disk space to testing that replication traffic is flowing.

A robust monitoring setup is essential for cluster health and viability – if your replicator goes offline and you do not know about it, then that slave becomes effectively useless because it has stale data.

Nagios Checks The Power of Persistence

One …