4 ms·

It is untrue that logs are trash. I gag at the comparison. If the would just look at logs they could answer so many questions that are costing a lot of money.

by worik 6y ago

It is untrue that logs are trash. I gag at the comparison. If the would just look at logs they could answer so many questions that are costing a lot of money. I am used to a Unix world, with logs as text files. They take up very little room, and it is easy to through out old ones (keeping logs for ever is foolish for a lot of reasons). I am staggered that "I turned logging on extensively for a recent Azure project. Some logs cost more than the service they were monitoring." There is no way that this can be true unless there is some very very bad craziness going on. A real mindfuck Even a hours worth of logs would of by now saved at least a week of developer time for the people I work with (my time, they pay for). "Logs are typically off by default in most Enterprise software" That is probably true, I have no reason to doubt you and it fits with the other things I have witnessed, but there is a better way.

jiggawatts 6y ago
> There is no way that this can be true unless there is some very very bad craziness going on. A real mindfuck System Center Operations Manager and all derived products such as the various Azure Monitor and Azure Logs offerings have craziness going on. For example, a typical metric is stored with a 300:1 write amplification because the database schema they use repeats the fully qualified machine name, counter name, and counter instance name per measurement. This prevents the collection of a useful set of metrics at a useful frequency. Logs are similarly generated using suuuuuuper verbose JSON, capturing everything except the critical fields you actually wanted, and then stored uncompressed record-by-record, keeping the full field names for each and every row. What incentive do you think they have to improve on the efficiency of the wire format or the storage format if they charge by the gigabyte?
- worik 6y ago
  Perhaps it is too much to ask for the HTTP request strings dumped to a file? If it is too much for production, turn it on for development servers. My foreign friends seem to know not what that means

by worik 6y ago

jiggawatts 6y ago

> There is no way that this can be true unless there is some very very bad craziness going on. A real mindfuck System Center Operations Manager and all derived products such as the various Azure Monitor and Azure Logs offerings have craziness going on. For example, a typical metric is stored with a 300:1 write amplification because the database schema they use repeats the fully qualified machine name, counter name, and counter instance name per measurement. This prevents the collection of a useful set of metrics at a useful frequency. Logs are similarly generated using suuuuuuper verbose JSON, capturing everything except the critical fields you actually wanted, and then stored uncompressed record-by-record, keeping the full field names for each and every row. What incentive do you think they have to improve on the efficiency of the wire format or the storage format if they charge by the gigabyte?

worik 6y ago
Perhaps it is too much to ask for the HTTP request strings dumped to a file? If it is too much for production, turn it on for development servers. My foreign friends seem to know not what that means