feat(server): Add support for aliasing commands

[abhijat]

Support for aliasing a command is added with the command line flag command_alias. Multiple aliases of the same command are allowed.

Example run:

server

./dragonfly --command_alias __set=set,doset=set

client

127.0.0.1:6379> __set foo bar
OK
127.0.0.1:6379> get foo
"bar"
127.0.0.1:6379> doset a b
OK
127.0.0.1:6379> get a
"b"
127.0.0.1:6379>

commands which have been renamed using rename_command should be aliased with the renamed command name instead of original.

TODO: update docs

[abhijat]

PR can be reviewed now but will wait for merge for 1.29

[kostasrim]

Mostly LGTM -- minor comments. For me I need to validate aliases and ping tomorrow.

[adiholden]

The goal of this task is to be able to filter commands from dashboard.
The suggested implementation is to use command alias. Please make sure that with your implementation when executing the command alias the metrics that are exposed to Prometheus are for the alias command string

[abhijat]

The goal of this task is to be able to filter commands from dashboard. The suggested implementation is to use command alias. Please make sure that with your implementation when executing the command alias the metrics that are exposed to Prometheus are for the alias command string

This is missing in this PR right now. Is there a list of metrics that are used to count the commands? I will make changes to update them for the alias.

[abhijat]

@adiholden after giving this more thought I suggest that this PR should be merged as it is (once reviewed and approved etc), it is self contained and adds the aliasing feature.

I can do a follow up PR to change metrics to be counted per alias or command.

I see the metrics with command name labels are

• commands_total
• commands_duration_seconds

Are there any other metrics to be updated?

[adiholden]

I can do a follow up PR to change metrics to be counted per alias or command.

Sure, I did not review the PR. I want you to think on the next step of the implementation and see that this PR will support this change need for metrics.

I see the metrics with command name labels are

• commands_total
• commands_duration_seconds

Are there any other metrics to be updated?

I believe that is all.

Also not sure what is the expected behaviour for info commandstats, maybe @romange can answer this

[adiholden]

I can do a follow up PR to change metrics to be counted per alias or command.

Sure, I did not review the PR. I want you to think on the next step of the implementation and see that this PR will support this change need for metrics.

I see the metrics with command name labels are

• commands_total
• commands_duration_seconds

Are there any other metrics to be updated?

I believe that is all.

Also not sure what is the expected behaviour for info commandstats, maybe @romange can answer this

Make sure you follow up on this and you know how you are going to implement the next step so this PR will supprot the change needed for the metrics

[kostasrim]

LGTM minor comments

[kostasrim]

Suggested change

	#include "absl/strings/str_split.h"
	#include <absl/strings/str_split.h>

[abhijat]

fixed

[kostasrim]

Shouldn't we expect that it->second will exist in the cmd_map_? In other words shouldn't this if be a DCHECK + a find. The registry is not mutable and the alias map is initialized during bootstrap so I would expect that this alias_lookup == cmd_map_.end() can't be true right ?

Basically what I am trying to say is that the alias map should never point to a command that does not exist in the cmd_map correct?

[abhijat]

The aliases are processed before commands are registered, in the ctor of command registry. It wouldn't be possible to validate if the alias is correct at that time.