Note

We are no longer onboarding new users to Citus Cloud on AWS. However, existing customers can continue to use Citus Cloud as they do today.

Monitoring

Cloud Platform Status

Any events affecting the Citus Cloud platform as a whole are recorded on status.citusdata.com. This page shows a chart of recent uptime, as well as listing any past incidents.

Resources Usage

Citus Cloud metrics enable you to get information about your cluster’s health and performance. The “Metrics” tab of the Cloud Console provides graphs for a number of measurements, all viewable per node. Below are the metrics, broken into categories, with details about the less obvious ones.

Amazon EBS Volume Metrics

(A full description of the metrics reported by Citus Cloud for underlying EBS resources can be found in the AWS documentation about I/O Characteristics and Monitoring.)

IOPS are a unit of measure representing input/output operations per second. The operations are measured in KiB, and the underlying drive technology determines the maximum amount of data that a volume type counts as a single I/O.

  • Read IOPS

  • Write IOPS

Volume queue length is the number of pending I/O requests for a device. Latency is the true end-to-end client time of an I/O operation, in other words, the time elapsed between sending an I/O and receiving an acknowledgement that the I/O read or write is complete.

  • Average Queue Length (Count)

  • Average Read Latency (Seconds)

  • Average Write Latency (Seconds)

  • Bytes Read / Second

  • Bytes Written / Second

CPU and Network

  • CPU Utilization (Percent)

  • Network - Bytes In / Second

  • Network - Bytes Out / Second

PostgreSQL Write-Ahead Log

  • WAL Bytes Written / Second

Formation Events Feed

To monitor events in the life of a formation with outside tools via a standard format, we offer RSS feeds per organization. You can use a feed reader or RSS Slack integration (e.g. on an #ops channel) to keep up to date.

On the upper right of the “Formations” list in the Cloud console, follow the “Formation Events” link to the RSS feed.

location of "new formation" button

The feed includes entries for three types of events, each with the following details:

Server Unavailable

This is a notification of connectivity problems such as hardware failure.

  • Formation name

  • Formation url

  • Server

Failover Scheduled

For planned upgrades, or when operating a formation without high availability that experiences a failure, this event will appear to indicate a future planned failover event.

  • Formation name

  • Formation url

  • Leader

  • Failover at

For planned failovers, “failover at” will usually match your maintenance window. Note that the failover might happen at this point or shortly thereafter, once a follower is available and has caught up to the primary database.

Failover

Failovers happen to address hardware failure, as mentioned, and also for other reasons such as performing system software upgrades, or transferring data to a server with better hardware.

  • Formation name

  • Formation url

  • Leader

  • Situation

  • Follower

Disk Almost Full

The “disk-almost-full” event happens if the disk utilization in the Postgres data directory reaches or exceeds 90%.

  • Formation name

  • Formation url

  • Server

  • “used_bytes”

  • “total_bytes”

  • “threshold_pct” (always 90)

The alert will appear once every twenty-four hours until the disk usage is resolved.

High CPU Utilization

The “cpu-utilization-high” event happens when there is CPU utilization reaches or exceeds 90%.

  • Formation name

  • Formation url

  • Server

  • “util_pct” (0-100% utilization value)

  • “threshold_pct” (always 90)

The alert will appear once per hour until CPU utilization goes down to a normal level.

StatsD external reporting

Citus Cloud can send events to an external StatsD server for detailed monitoring. Citus Cloud sends the following statsd metrics:

Metric

Notes

citus.disk.data.total

citus.disk.data.used

citus.load.1

Load in past 1 minute

citus.load.5

Load in past 5 minutes

citus.load.15

Load in past 15 minutes

citus.locks.granted.<mode>.<locktype>.count

See below

citus.mem.available

citus.mem.buffered

citus.mem.cached

citus.mem.commit_limit

Memory currently available to be allocated on the system

citus.mem.committed_as

Total amount of memory estimated to complete the workload

citus.mem.dirty

Amount of memory waiting to be written back to the disk

citus.mem.free

Amount of physical RAM left unused

citus.mem.total

Total amount of physical RAM

citus.pgbouncer_outbound.cl_active

Active client connections

citus.pgbouncer_outbound.cl_waiting

Waiting client connections

citus.pgbouncer_outbound.sv_active

Active server connections

citus.pgbouncer_outbound.sv_idle

Idle server connections

citus.pgbouncer_outbound.sv_used

Server connections idle more than server_check_delay

citus.postgres_connections.active

citus.postgres_connections.idle

citus.postgres_connections.unknown

citus.postgres_connections.used

Notes:

  • The citus.mem.* metrics are reported in kilobytes, and are also recorded in megabytes as system.mem.*. Memory metrics come from /proc/meminfo, and the proc(5) man page contains a description of each.

  • The citus.load.* metrics are duplicated as system.load.*.

  • citus.locks.granted.* and citus.locks.not_granted.* use mode and locktype as present in Postgres’ pg_locks table.

  • See the pgBouncer docs for more details about the pgbouncer_outbound metrics.

To send these metrics to a statsd server, use the “Create New Metrics Destination” button in the “Metrics” tab of Cloud Console.

metrics tab showing create metrics destination button

Then fill in the host details in the resulting dialog box.

screen to enter details of destination

The statsd protocol is not encrypted, so we advise setting up VPC peering between the server and your Citus Cloud cluster.

Example: Datadog with statsd

Datadog is a product which receives application metrics in the statsd protocol and makes them available in a web interface with sophisticated queries and reports. Here are the steps to connect it to Citus Cloud.

  1. Sign up for a Datadog account and take note of your personal API key. It is available at https://app.datadoghq.com/account/settings#api

  2. Launch a Linux server, for instance on EC2.

  3. In that server, install the Datadog Agent. This is a program which listens for statsd input and translates it into Datadog API requests. In the server command line, run:

    1. # substitute your own API key
    2. DD_API_KEY=1234567890 bash -c \
    3. "$(curl -L https://raw.githubusercontent.com/DataDog/datadog-agent/master/cmd/agent/install_script.sh)"
  4. Configure the agent. (If needed, see Datadog per-platform guides)

    1. cat - | sudo tee -a /etc/datadog-agent/datadog.yaml << CONF
    2. non_local_traffic: yes
    3. use_dogstatsd: yes
    4. dogstatsd_port: 8125
    5. dogstatsd_non_local_traffic: yes
    6. log_level: info
    7. log_file: /var/log/datadog/agent.log
    8. CONF
    9. # this is how to do it on ubuntu
    10. sudo systemctl restart datadog-agent
  5. Fill in the agent server information as a new metrics destination in the Cloud Console. See the previous section for details.

  6. The agent should now appear in the Infrastructure section in Datadog.

    hostname of service, with status "UP"

Clicking the hostname link goes into a full dashboard of all the metrics, with the ability to write queries and set alerts.

VividCortex External Monitoring

Like the systems above, VividCortex provides a metrics dashboard. While the other systems mostly focus on computer resources, VividCortex focuses on the performance of queries. It tracks their throughput, error rate, 99th percentile latency, and concurrency.

To integrate VividCortex with Citus Cloud we’ll be using the Off-Host Configuration. In this mode we choose a database role with permissions to read the PostgreSQL statistics tables, and give the role’s login information to the VividCortex agent. VividCortex then connects and periodically collects information.

Here’s a step-by-step guide to get started.

Note

The steps below give VividCortex access to the database as the “citus” role. This role has fairly elevated permissions. Consider possible issues of data confidentiality before enabling VividCortex monitoring.

  1. Create a special VividCortex schema and relations on the Citus coordinator node.

    1. # Use their SQL script to create schema and
    2. # helper functions to monitor the cluster
    3. curl -L https://docs.vividcortex.com/create-stat-functions-v96.sql | \
    4. psql [connection_uri]
  2. Create a VividCortex account.

  3. The Summary page will prompt you to “Install Database Performance Monitor On A New Host.” Choose OFF-HOST.

    installation methods selector

  4. Select the PostgreSQL database.

    database selector

  5. Input the connection information into the credentials screen in the VividCortex wizard. The “citus” user is required. Make sure SSL Enabled is on, and that you’re using SSL Mode “Verify Full.” Specify /etc/ssl/certs/citus.crt for the SSL Authority.

    database connection details

  6. Skip the create user step. We’ll be using the “citus” user, because it’s the only one with enough permissions to view query statistics on Citus Cloud.

    ../_images/vc-create-user.png

    (Skip this step)

  7. Skip the configuration step. We know that pg_stat_statements is pre-configured properly on Citus Cloud.

    ../_images/vc-config.png

    (Skip this step)

  8. Provision a server to act as the VividCortex agent. For instance a small EC2 instance will do. On this new host install the Citus Cloud SSL certificate.

    1. sudo curl -L https://console.citusdata.com/citus.crt \
    2. -o /etc/ssl/certs/citus.crt
  9. Advance to the next screen in the wizard. It will contain commands to run on the agent server, customized with a token for your account.

    textbox with commands to copy

    After running the commands on your server, the server will appear under Select host. Click it and go to the next step in the wizard.

    confirmation that host is active

  10. After these steps, VividCortex should show all systems as activated.

checklist of all components active

You can then proceed to the dashboard to monitor queries on your Citus cluster.

live bar charts on dashboard