At Rockset, we regularly search for methods to provide our clients higher visibility into the product. Towards this aim, we not too long ago determined to enhance our customer-facing question logging. Our earlier iteration of question logs was primarily based in one among our shared providers known as apiserver. As a part of the work that apiserver would do when finishing a question execution request, it might create a log that might finally be ingested into the _events assortment. Nevertheless, there have been points that made us rethink this implementation of question logs:

1. No isolation: as a result of the question logs in _events relied on shared providers, heavy site visitors from one org may have an effect on question logging in different orgs.
2. Incomplete logs: due to the problems prompted by utilizing shared providers, we solely logged question errors – profitable queries wouldn’t be logged. Moreover, it was not potential for us to log information about async queries.
3. No means to debug question efficiency – the question logs in _events solely contained fundamental details about every question. There was no means for the person to get details about why a given question could have run slowly or exhausted compaute sources for the reason that logs contained no details about the question plan.

Improved Question Logging

The brand new question logs function addresses all of those points. The mechanisms that deal with question logs are contained solely inside your Digital Occasion versus being inside one among Rockset’s shared providers. This offers question logs the benefit of isolation. Moreover, each question you submit shall be robotically logged you probably have already created a set with a question logs supply (offered you don’t hit a price restrict).

How Question Logs Work

Question logging begins on the finish of question execution. As a part of the steps which can be run within the closing aggregator when a question has accomplished, a document containing metadata related along with your question is created. At this level, we might also have to gather info from different aggregators that have been concerned within the question. After that is performed, the document is briefly saved in an in-memory buffer. The contents of this buffer are flushed to S3 each few seconds. As soon as question logs have been dumped to S3, they are going to be ingested into any of your question log collections which have been created.

INFO vs DEBUG Logs

After we first designed this challenge, we had at all times meant for it to work with the question profiler within the console. This is able to enable our clients to debug question bottlenecks with these logs. Nevertheless, the question profiler requires fairly a bit of knowledge, that means it might be not possible for each question log to comprise all the knowledge essential for the profiler. To resolve this downside, we opted to create two tiers of question logs – INFO and DEBUG logs.

INFO logs are robotically created for each question issued by your org. They comprise some fundamental metadata related along with your question however can’t be used with the question profiler. When you recognize that you could be wish to have the power to debug a sure question with the profiler, you’ll be able to specify a DEBUG log threshold along with your question request. If the question execution time is bigger than the desired threshold, Rockset will create each an INFO and a DEBUG log. There are two methods of specifying a threshold:

1. Use the debug_log_threshold_ms question trace

   SELECT * FROM _events HINT(debug_log_threshold_ms=1000)

2. Use the debug_threshold_ms parameter in API requests. That is accessible for each question and question lambda execution requests.

Observe that since DEBUG logs are a lot bigger than INFO logs, the speed restrict for DEBUG logs is way decrease. For that reason, it’s endorsed that you just solely present a DEBUG log threshold when you recognize that this info could possibly be helpful. In any other case, you run the chance of hitting the speed restrict whenever you most want a DEBUG log.

System Sources

As a part of this challenge, we determined to introduce a brand new idea known as system sources. These are sources which ingest information originating from Rockset. Nevertheless, in contrast to the _events assortment, collections with system sources are managed solely by your group. This lets you configure all the settings of those collections. We shall be introducing extra system supply varieties as time goes on.

Getting Began with Question Logging

To be able to begin logging your queries, all you’ll want to do is create a set with a question logs supply. This may be performed by means of the console.

Rockset will start ingesting question logs into this assortment as you submit queries. Logs for the final 24 hours of queries may also be ingested into this assortment. Please notice that it could possibly take a couple of minutes after a question has accomplished earlier than the related log will present up in your assortment.

To be able to use the question profiler with these logs, open the Rockset Console’s question editor and difficulty a question that targets one among your question logs collections. The question editor will detect that you’re making an attempt to question a set with a question logs supply and a column known as ‘Profiler’ shall be added to the question outcomes desk. Any paperwork which have a populated stats area can have a hyperlink on this column. Clicking on this hyperlink will open the question profile in a brand new tab.

Observe that customized ingest transformations or question aliases can intervene with this performance so it’s endorsed that you don’t rename any columns.

For an intensive dive into utilizing Rockset’s Question Profiler, please discuss with the video accessible right here.

Conclusion

Hopefully, this has given you a fast look into the performance that question logs can supply. Whether or not you’ll want to debug question efficiency or test why beforehand accomplished queries have failed, your expertise with Rockset shall be improved by making use of question logs.