It's simply great, well indexed by Google and has meaningful human-readable URLs, so one can even guess them for the topics he need. I have http://dev.mysql.com/doc/refman/5.6/en/merge-storage-engine.html open right now and looking at it I clearly understand without any search that if I need a reference for SELECT syntax in 5.5 I have just replace "5.6" with "5.5" and "merge-storage-engine" with "select" to get there. It works that way and many users even had created some nice marcos or shortcuts for their browser to end up at the proper manual page after typing just the topic they are interested it.
These being said, today I'd like to write about problems with MySQL manual that I care about. Some of them are not new and are hard to fix/avoid, others I consider recent enough. Today I'd like to discuss 4 of them:
- MySQL Manual does NOT have enough careful readers who are ready to report problems or missing details.
- Old documentation bug reports seem to be ignored without any obvious good reason.
- Some documentation pages are useless with their current content.
- "How to" kind of official MySQL documentation is missing, and this seems intentional.
Also, it's easy to understand that authors who had written the text are not the best people to read it later with a hope to find something useful or some mistake there. They deal with their "ideas" and may have too many things assumed in mind comparing to average user/reader of the manual. They are also very experienced MySQL users and even developers in the past. I had written a book myself long time ago and translated dozens of technical books into Russian, so I know the feeling - for you, as a writer, your text is clear, useful and may be just great. While it is NOT, especially for readers with different technical or cultural background.
As a result we have manual pages that are less than clear and has a lot of useful things missing or improperly documented. When I started to read the manual carefully some time ago (after I stopped writing about MySQL bugs at Facebook I needed some other thing to do to keep my mind in peace) I had found out that I can easily report a bug (if not 5 or more) for any page I open. I see missing details or wrong statements everywhere. I had even "invented" separate "missing manual" tag to classify some of the documentation bugs I've reported. It's so easy to find these problems and report them as bugs in "Server: Documentation" category...
So, MySQL manual has bugs, but not so many it seems (I see 86 active ones at the moment). This is NOT because it's ideal, but more because too few people care to read the manual carefully and report anything they consider wrong or missing. Actually, besides MySQL Support engineers and few other Oracle employees, few of my colleagues at Percona and myself, I know only 3 Community members who report documentation bugs recently:
- Daniel van Eeden. He is a well known Community bug reporter and used to be great Oracle customer (at least this is how I remember him from the days when I worked in Oracle MySQL Support). Out of his 78 active bug reports I see 11 for MySQL Server manual, but he also reports bugs in documentation for Oracle's "commercial" MySQL software, like MySQL Enterprise Backup.
- Peter Laursen. He is also a well known and productive MySQL bugs reporter for a decade probably. Out of his 83 active bug reports 5 are related to MySQL Manual. They are not very recent though.
- Federico Razzoli. He had started to report himself only recently because of privacy-related issues with information requested by Oracle to obtain SSO account for bugs reporting. Before that he asked me to report bugs sometimes, and I did. So, now of his 5 active bugs 2 are related to MySQL Manual.
Other important problem of the manual related to bugs is the fact that some of the bugs reported are NOT fixed in time. Let me share my favorite example here, my Bug #68097. It was reported almost 14 months ago and asked to actually properly document PERFORMANCE_SCHEMA behavior that was intended and explained long time ago - the only way to enable instrumentation for (InnoDB) rwlocks and mutexes is upon server startup. If these instruments where not enabled at startup they are just not available, otherwise you can disable them and enable them back dynamically. I fail to understand what's so hard to write exactly this, as we can find statements about this from Oracle engineers in many places.
As a result, MySQL Manual stays misleading for years! By the way, the oldest documentation bug I see in "Verified" status, Bug #30538, remains in this state without further comments since October 19, 2010. Based on recent comments it may be even somehow fixed, but nobody cares to change status then. And this is for a bug reported back in 2007 by famous customer (probably Google at that time), Mark Callaghan. Obviously we see not enough care about older documentation requests. My recent ones are usually fixed fast though.
Some pages of MySQL Manual exist for years, but their content is almost useless without looking into the code or checking with developers. Great examples can be found in documents devoted to NDB Cluster internals, for example, this page (it's one of many): https://dev.mysql.com/doc/ndbapi/en/ndb-internals-dump-command-2401.html. Nobody even tried to explain the output here, interpretation of the fields, use cases for this DUMP command, nothing. Just few words and example of output that I can probably get myself. How this page is useful to anybody?
Unfortunately, the problem is NOT limited to NDB Cluster internals (that I am happy to honestly not care about at all at the moment). Same happens even to InnoDB. Check http://dev.mysql.com/doc/refman/5.6/en/innodb-monitors.html and tell me where it explains the details about any lock waits related information you can find out in TRANSACTIONS section of INNODB STATUS output. I see only this text:
"If this section reports lock waits, your applications might have lock contention. The output can also help to trace the reasons for transaction deadlocks."That's all! Really. It does not even explain how "next-key lock" looks in the output vs "record lock"... As a result, one can not explain the way to troubleshoot even simple deadlock based on the manual or, if she tries, she can NOT prove the points with anything but experiments and quotes from the source code. And all that happens when InnoDB lock model does NOT change for decade(s) and source code seems commented in details.
The fourth problem, the lack of official "How to" or "FAQ" documents, is closely related to the above. Not only MySQL Manual sometimes misses a lot of details, you can not get them from http://dev.mysql.com/doc/ (the place to look for official MySQL guides) at all! How to read the output of SHOW ENGINE INNODB STATUS? Now to use PERFORMANCE_SCHEMA to troubleshoot different types of performance problems? You can find answers (of different quality, level of details and age) in great blog posts, books and presentations, maybe in Oracle's knowledge base articles (if you are lucky to be their customer and good in search there) but NOT in official MySQL documentation.
I do not consider this normal or acceptable, especially for the open source software. Check FreeBSD Documentation Project, for example, or Linux Documentation Project to get some ideas of how other open source software approaches the problem.
Surely, adding a collection of "How to" guides probably requires input from Community, but why at least not to try to collect links to all the good existing documents of this kind somewhere at http://dev.mysql.com? Having third party documents of this kind with "officially approved" status after a review by Oracle's MySQL Documentation and/or Support teams would help a lot, I think.
That's all I can complain about at the moment. One day I'll try to continue...