This post is part of a bigger topic Autoscaling Publishers in AWS.
In order to define autoscaling of some servers, we need some metrics that we can use to create the autoscaling logic, i.e. when to spin up new instances and when to terminate them. A good measure for this, in Tridion terms, is the size of the Publishing Queue. Namely for Publishers autoscaling, it's useful to look at the number of items in the Publishing Queue that are in the state "Waiting for Publish".
The approach is to read this metric somehow from the Tridion Content Manager database and make it available in AWS, so that we can use it later. AWS CloudWatch provides a way to define and/or intercept events that can trigger some code execution. The code executed is supposed to read the Publishing Queue and push the count of items into CloudWatch as a custom metric.
After a few iterations and optimizations, the code looks like this:
In order to define autoscaling of some servers, we need some metrics that we can use to create the autoscaling logic, i.e. when to spin up new instances and when to terminate them. A good measure for this, in Tridion terms, is the size of the Publishing Queue. Namely for Publishers autoscaling, it's useful to look at the number of items in the Publishing Queue that are in the state "Waiting for Publish".
The approach is to read this metric somehow from the Tridion Content Manager database and make it available in AWS, so that we can use it later. AWS CloudWatch provides a way to define and/or intercept events that can trigger some code execution. The code executed is supposed to read the Publishing Queue and push the count of items into CloudWatch as a custom metric.
1. Define Lambda Function
This function represents the code that is executed by the CloudWatch rule. The function reads the size of the Publishing Queue and pushes it as custom metrics into CloudWatch.
The languages available in AWS Lambda at the moment include .Net Core 1 and Python 2.7. I tried writing a nice .net application that uses Tridion's CoreService client to read the Publishing Queue metrics I needed. Unfortunately, I had to give this up after realizing the limitations in .Net Core 1 regarding connectivity to WCF services. Connecting to a service is really a big deal in 2017 -- you need a ton of DLLs!
Instead, I wrote the Lambda code in Python 2.7 using direct DB access to read the metrics from the Tridion CM DB. Definitely not the nicest approach, but it seems like the only way to do it. Also because the DB is an RDS instance in the same VPC, I wasn't too concerned with security.
After a few iterations and optimizations, the code looks like this:
from os import getenv import pymssql import boto3 client = boto3.client('cloudwatch') def handler(event, context): server = getenv("PYMSSQL_SERVER") user = getenv("PYMSSQL_USERNAME") password = getenv("PYMSSQL_PASSWORD") database = getenv("PYMSSQL_DB") conn = pymssql.connect(server, user, password, database) cursor = conn.cursor() cursor.execute('select STATE, COUNT(*) from PUBLISH_TRANSACTIONS where STATE=1 or STATE=4 group by STATE') metrics = {'Waiting for Publish': 0, 'Waiting for Deployment': 0} for row in cursor.fetchall(): count = row[1] if row[0] == 1: metrics['Waiting for Publish'] = count elif row[0] == 4: metrics['Waiting for Deployment'] = count print 'Metrics', metrics for metric in metrics: response = client.put_metric_data( Namespace='SDL Web', MetricData=[ { 'MetricName': metric, 'Value': metrics[metric], 'Unit': 'Count', }, ] ) conn.close()
I used environment variables in order to make the code more portable and clean. These variables are specified in the AWS console.
The code reads 2 values:
Since I'm going to implement autoscaling for Deployers, I might as well read the relevant metrics in one go.
The code reads 2 values:
- number of items in Waiting for Publish state;
- number of items in Waiting for Deployment state;
Since I'm going to implement autoscaling for Deployers, I might as well read the relevant metrics in one go.
The code uses pymssql library to interact with CM DB. It also uses the boto3 CloudWatch client to push the custom metrics into CloudWatch.
2. Define Rule in CloudWatch
CloudWatch rules can be defined based on a time schedule (like a cron job) or based on events raised somewhere else.
In this situation, a time pattern rule made sense. So I created a rule that fires every minute.
You also associate a target with the rule. This specifies what happens when the rule fires. In my case it executes the Lambda function created in step 1.
Give the rule a name and a description and save it.
3. Visualize Data in Dashboard
One can inspect the new custom metrics in CloudWatch and use them for creating alarms (presented in a later post) or place them in a dashboard like this:
Comments