feat(correlation-id) new "tracker" generator

[WALL-E]

SEE ALSO #1228

[Tieske]

👍 great! code lgtm

What's left;

• update docs
• adding a test

[thibaultcha]

Could we cache all those global accesses? Something like:

local fmt = string.format
local now = ngx.now
local pid = ngx.worker.pid -- only need to retrieve it once

-- ...

local var = ngx.var
return fmt(...,
var.server_addr,
var.server_port,
pid,
-- ...
now()
)

[thibaultcha]

I'm afraid the chances of collision is really high with this generator. ngx.now sure provides milliseconds precision but surely that won't be enough to guarantee a level of uniqueness as high as uuids.

[WALL-E]

updated doc is done.

[Tieske]

I'm afraid the chances of collision is really high with this generator. ngx.now sure provides milliseconds precision but surely that won't be enough to guarantee a level of uniqueness as high as uuids.

yes, crossed my mind as well. But figured it would be easy to add a counter within the millisecond precision if needs be and it actually causes collisions. But maybe better to prevent it any way and add such a counter right away.

[WALL-E]

in modern operating system, pid_t is int type, and the pid will be increasing(pid%MAX_INT),
at the same time(milliseconds precision), I think the pid not repeat.

can you tell me that I'm wrong? @Tieske @thibaultcha

[Tieske]

not sure, but I don't think you should be caching the magic table ngx on module level. Sen t you an update.

Not sure I understand your comment (with pid_t) but the question imo is whether we'll assume there will be no collisions, or we code to make sure there are none.

[timerickson]

And dont forget subsequent requests may come to different nodes on different hosts who may have varying clocks which could increase the chances of collision.

[WALL-E]

Correlation ID = "ip-port-pid-connection-connection_requests-timestamp"

ip: different hosts
port: different kong node
pid: different process (pid's type is pid_t, this number can be very large when increasing)
connection: serial number (different connection)
connection_requests: serial number on same connection (different request when HTTP Keepalive)
timestamp: milliseconds precision (different time)

code can be make sure there are no collisions,
but I'm not familiar with kong's cluster specifics, I'm not sure it is ok in the cluster.

[thibaultcha]

I see you cached string.format, but please do follow my other caching suggestions.

[Tieske]

I already gave it a quick edit; https://github.com/WALL-E/kong/pull/1

[thibaultcha]

Right, the combination of $connection and $connection_requests should avoid any collision indeed. This is still missing a test.

[Tieske]

typo: generator

[Tieske]

@WALL-E can you rebase on next? LGTM, except for the typo.

@thibaultcha any more comments, or ready to merge? (after rebase)

[thibaultcha]

Manually applied the patch with minor modifications to next. Thank you for your contribution!

[kamoljan]

Looking forward to next release :)

[WALL-E]

I seem to have missed the beautiful moment :(