gin | A high performance C playground library | Performance Testing library

Like a_horse already advised (and you mentioned yourself), the proper fix is to extract those attributes to separate columns, normalizing your design to some extent.

Sadly, no (as of Postgres 14).

It could work in theory. Since your values are big, an expression index with just some small attributes can be picked up by Postgres in an index-only scan, even when retrieving all rows (where it otherwise would ignore indexes).

The manual:

However, PostgreSQL's planner is currently not very smart about such cases. It considers a query to be potentially executable by index-only scan only when all columns needed by the query are available from the index.

So you would have to include value itself in the index, even if just as INCLUDE column - totally spoiling the whole idea. No go.

You probably can still do something in the short term. Two crucial quotes:

I am looking to see if I can make this faster in short term

The json blobs are a bit large

Drop the cast to json from the query. Casting every time adds pointless cost.

One major cost factor will be compression. Postgres has to "de-toast" the whole large column, just to extract some small attributes. Since Postgres 14, you can switch the compression algorithm (if support is enabled in your version!). The default is dictated by the config setting default_toast_compression, which is set to pglz by default. Currently the only available alternative is lz4. You can set that per column. Any time.

LZ4 (lz4) is considerably faster, while compressing typically a bit less. About twice as fast, but around 10 % more storage (depends!). If performance is not an issue it's best to stick to the stronger compression of the default LZ algorithm (pglz). There may be more compression algorithms to pick from in the future.

To implement:

Using middlewares is certainly not the way to go here, your hunch is correct! Using FastAPI as inspiration, I usually create models for every request/response that I have. You can then bind these models as query, path, or body models. An example of query model binding (just to show you that you can use this to more than just json post requests):

Better answer: as suggested by @Oliver Dain, use zap.AtomicLevel. See their answer for details.

Another option is to create a core with a custom LevelEnabler function. You can use zap.LevelEnablerFunc to convert a closure to a zapcore.LevelEnabler.

The relevant docs:

LevelEnabler decides whether a given logging level is enabled when logging a message.

LevelEnablerFunc is a convenient way to implement zapcore.LevelEnabler with an anonymous function.

That function may then return true or false based on some other variable that changes at runtime:

The flag gin.DebugMode is used to control the output of gin.IsDebugging(), which adds some additional log output and changes the HTML renderer to the debug struct HTMLDebug.

The gin.TestMode is used in Gin's own unit tests to toggle the debug mode (and the additional logging) on and off, and the usage of the debug HTML renderer.

Other than that, it doesn't have other uses (source).

However, the flag can be controlled with the environment variable GIN_MODE=test. Then, since Mode() is exported, you can use it in application code to, for example, declare testing routes. This might have some merit if you plan to run an E2E test suite, or some other integration test:

You need to add WorkingDirectory to your system file

Here's one idea to help visualize the difference you are trying to show. Firstly, the country names on the x axis are likely to remain illegible however you try to label them. It might therefore be better to have the rank of the countries on the x axis.

Drawing a polygon around the points might make the point visually, but doesn't make much sense in statistical terms. What might be better here is to plot a regression with a separate line for each group. Since we are dealing with count data, we can use Poisson regression, and since we have a numeric rank on the x axis, it is possible to have lines going across your plot to show the regression.

Create a net.Listener implementation that accepts connections by receiving on a channel:

You should have those two indexes to speed up your query :

Use a sub-engine.

You can instantiate an engine with gin.New for internal routes only and not run it. Instead you pass the context from your Any route to Engine.HandleContext.

This will relay the context from the main engine, and it will match path params based on the placeholders in the sub-routes.

You can declare routes on the sub-engine as usual: