MySQL的FULLTEXT搜索是否为MyISAM和InnoDb返回相同的结果？

Question

If you take a table and simply change the storage engine engine from MyISAM to InnoDb will all WHERE MATCH (col1,col2,col3...) AGAINST (expr) return exactly the same results as under MyISAM? If no, what are the differences?

I mean just differences in terms of fulltext searches, nothing else. There are obviously other huge differences in these two storage engines.

Answer 1

There are actually some notable differences in the implementation of the MyISAM and InnoDB fulltext searches:

• The MyISAM natural language search (but not the boolean mode) has a 50% threshold, while InnoDB doesn't, so very (very) common words are excluded in the MyISAM results. There is a remark in the manual about it:

  The 50% threshold can surprise you when you first try full-text searching to see how it works, and makes InnoDB tables more suited to experimentation with full-text searches. If you create a MyISAM table and insert only one or two rows of text into it, every word in the text occurs in at least 50% of the rows. As a result, no search returns any results until the table contains more rows.

• The MyISAM stopword list (a list of words that is not included in the fulltext index and thus cannot be found) is significantly longer than the (default) one used by InnoDB, so eg "everybody" or "unfortunately" can be found with InnoDB, but not with MyISAM. match against ('Mary Had a Little Lamb') will usually contain a lot more results, as "had" is a stopword in MyISAM, but not in InnoDB.

• MyISAM and InnoDB use different weight algorithms. MyISAM considers eg the ratio of matching words to non-matching words in a row, so a long sentence that contains a word is less relevant than a short sentence with that word. Although this will only change the order in the otherwise identical resultset, this oftentimes has a significant impact on the user experience and if the user regards two results as "the same" , which is what you are asking about. This might also be particular relevant, as searches usually include a limit, eg order by score desc limit 10 , which thus can yield completely different results.

• InnoDB supports "" to match exact phrases (words in given order), while MyISAM (at least in natural language mode) doesn't. So if you use match against ('"Mary Had a Little Lamb"') , InnoDB will only return a row if it contains this exact sentence, while MyISAM will find every row that contains any of these words (apart from "had" as mentioned above, and "a", which is in both stopword lists).

• Since you are using the natural language mode, deviations in the boolean search are probably not relevant for you, but to list at least one: the two engines differ in how they treat stop (or short) words in the search query. If you use match against ('+about +Mary' in boolean mode) ("about" is a stopword in both engines), InnoDB will try to find that word in the index although it cannot be in there, and thus return no results, while MyISAM will ignore that word and can return results that may not contain "about", only "Mary".

Additionally, the default values for the minimum word length, ft_min_word_len for MyISAM (default 4) and innodb_ft_min_token_size for InnoDB (default 3) are different, so if you do not adjust them, the InnoDB index will contain (and find) more words. You might also want to adept the stop word list to match each other.

If these differences are relevant in your case will depend on your data, your search patterns and if you consider a different order to be a different result. Searches in data that mainly consists of short terms or fixed formats, eg product codes or company names, or searches where you are mainly interested in finding specific words at all, or searches that usually only yield a handful of possible results, will usually vary less in the two engines than searches in actual english texts, where a different relevance score has a bigger effect.

Answer 2

No, there's no guarantee that an InnoDB fulltext index works exactly the same way as the MyISAM fulltext index on the same data.

Last time I tested it (which was when InnoDB FT was still Beta), there were definitely cases where InnoDB FT did not return some rows that matched in MyISAM FT. It also returned some rows that were not matched in MyISAM.