too many requests to CloudWatch

[yegor256]

We've sent 18 million requests to AWS CloudWatch over the last 30 days. This is too many.. They simply cost too much. Let's find a way to send less. Maybe we should collect data locally (in files) and send only hourly summaries to CW.

[davvd]

I'll ask someone to take care of this task soon

[davvd]

@yegor256 thanks for the ticket, your account was topped for 15 mins, payment 000-cc24bf87

[davvd]

@carlosmiranda it's your task now, the budget is 30 mins, go ahead

[carlosmiranda]

@yegor256 , for us to accomplish this we'd need to redesign the way the system is sending data to CloudWatch. At the very least, I'm thinking that we need to do the following:

1. DefaultHost should locally log data for each request instead of sending data directly to CloudWatch.
2. Create a scheduled job that consolidates the logs, makes a request to CloudWatch (one request per domain), and cleans up after.

It's a pretty big job and I suggest we do 1) first, and continue with the rest with PDD. For that, I'd ask, what format should the data be stored, and where?

[yegor256]

Yes, I agree. Let's use H2 database in a directory?

[yegor256]

@carlosmiranda it's an urgent problem, would be great if you can focus on it

[carlosmiranda]

@yegor256 , sure, sorry, I've got my hands full lately. Could you give me a couple of days?

Also, just to make things quick, here was what I was planning to implement so that you can review/critique:

1. An interface DomainStatsData, with methods void put(Domain, Stats) that will log the stats for each domain, including a timestamp, Stats get(Domain) that will get stats for each domain, and Map<Domain, Stats> get() for getting stats for all domains.
2. An implementation H2StatsData that uses an H2 file-based database. It will contain a static DataSource field which will be used for the read/write operations. We will write to a database /domainStats
3. DefaultResource will post data DomainStatsData instead of AmazonCloudwatch in DefaultResource
4. Puzzles will be left for the implementation of the batch job that will periodically send data to Amazon Cloudwatch.

What do you think?

[yegor256]

@carlosmiranda sounds perfect, please go ahead

[carlosmiranda]

I created a pull request for it at #177, please take a look.

[carlosmiranda]

@yegor256 #177 is now merged to master, are we good here?

[yegor256]

yes, many thanks!

[davvd]

@carlosmiranda Much appreciated! 30 mins was added to your account, payment ID 40929402

[davvd]

puzzles in progress: 185-1d178b38/#192

[davvd]

last puzzled solved 185-1d178b38/#192