internal/function: use cache for expensive language operations

[erizocosmico]

Closes #442

This adds a two queue lru cache to language UDF. Why a 2Q cache instead of a regular LRU? Because we do not only want the most recent but the most frequent as well to maximize the cache usage.

Since running enry on a file_path only is just marginally slower than using the cache, I decided to not use the cache unless the blob_content is being provided (we could use two caches, though) to not pollute the cache and use it only for the really expensive operations: reading stuff from blobs.

Note that the cache is per gitbase instance.

Results are pretty much the ones @ajnavarro describes in the issue.

Before:

mysql> select count(*) from (select file_path, language(file_path, blob_content) as language FROM files WHERE language(file_path, blob_content) = 'Go') t;
+----------+
| COUNT(*) |
+----------+
|    39781 |
+----------+
1 row in set (20.85 sec)

Now:

mysql> select count(*) from (select file_path, language(file_path, blob_content) as language FROM files WHERE language(file_path, blob_content) = 'Go') t;
+----------+
| COUNT(*) |
+----------+
|    39781 |
+----------+
1 row in set (17.14 sec)

[kuba--]

Right now, we are trying to let apps adjust cache sizes (and pass them down to packages like go-git), so maybe in this case we can also think about how easily we can modify this cache in the future (at least size).

[erizocosmico]

Yep, this should be configurable, right @ajnavarro? Environment variable or CLI flag that passes down to here? WDYT?

[ajnavarro]

yep, it should be configurable. it should be a env var o cli flag, yep.

[erizocosmico]

Added configuration to the cache size via env var

[kuba--]

👍

[ajnavarro]

shall we add the default value here?

[ajnavarro]

should we add that env variable as we are doing with others? check https://github.com/src-d/gitbase/blob/master/cmd/gitbase/command/server.go#L50

[erizocosmico]

If we add it there we will have to pass it all the way down until the infinite. Besides, there are other examples of environment variables such as blob max size that are not there but where they're used

[erizocosmico]

Added default value to docs