SMD HTTP client logs are huuuuuuuuge and always on

[rainest]

During testing on a small(ish) system, I noticed that we frequently check SMD for inventory endpoints, and that the resulting debug log for this is quite large:

2025/07/30 20:56:06 [DEBUG] POST http://smd/hsm/v2/State/Components/Query
2025/07/30 20:56:06 [DEBUG] GET http://smd/hsm/v2/Inventory/ComponentEndpoints?id=x1403c0s6&id=x1403c7s7b0&id=x1402c6s5b0n0&id=x1402c3s5b1n1&id=x1402c3r3b0&id=x1403c1s7&id=x1402c3s4e1&id=x3120m1p0v13&id=x1402c0s1b1n0&id=x1403c0s2e0&id=x1403c0s1b0&id=x1402c2s2b0n0&id=x1403c3s5&id=x1403c3s0&... about 20k char of every xname in the system

this takes up my whole screen and isn't particularly useful. I'd like to get rid of it, but this is pretty difficult with our current logger and HTTP client, as it's baked into the latter: https://github.com/hashicorp/go-retryablehttp/blob/v0.7.7/client.go#L658-L665

There's no way to ask for a truncated URL in output, but we should be able to use a logger that has satisfies their LeveledLogger interface. We could alternately provide no logger to retryablehttp and only perform our own logs above. Not sure how much else that'd lose us.

[rainest]

I realized that we don't actually have configurable log levels in PCS yet, and that's probably another case where CSM-inherited services are quite inconsistent (IMO migrate them to https://pkg.go.dev/go.uber.org/zap). Removed "good first issue" since that kinda scope creeps this a lot.

There's still the simpler option where we don't use the HTTP client logging and just perform our own outside it.

[alexlovelltroy]

In other parts of the code, we've introduced a chi middleware for consistent request logging. It uses zerolog rather than zap, but aims for the kind of thing you're looking for.

• Logging and Events roadmap#49
• https://github.com/OpenCHAMI/chi-middleware/blob/main/log/middleware.go

[rainest]

Zerolog's probably good enough, though I don't have experience with it.

This is for outbound requests rather than anything that goes through Chi though.

You can probably create child routers off the Zerolog instance and pass them or a HashiCorp interface wrapper to the outbound client fine, just the same thing where PCS doesn't have any consistent logging anywhere yet.

[davidallendj]

Is this related to how SMD is doing logging or only external to it? If so, shouldn't this issue be in SMD repo?

[rainest]

External. power-control generates these logs from its HTTP client when it sends requests to SMD.

power-control/internal/hsm/state_manager.go

Lines 368 to 393 in 0046c41

	idCount := numComps
	if idCount > b.HSMGlobals.MaxComponentQuery {
	idCount = b.HSMGlobals.MaxComponentQuery
	}
	//Construct the URL
	urlSuffix := ""
	for ix := 0; ix < idCount; ix++ {
	urlSuffix = urlSuffix + "&id=" + xnames[compIX]
	compIX++
	}
	smurl := b.HSMGlobals.SMUrl + hsmInventoryComponentEndpointPath +
	"?" + strings.TrimLeft(urlSuffix, "&")
	//Make the HSM call
	req, err := http.NewRequest(http.MethodGet, smurl, nil)
	if err != nil {
	return fmt.Errorf("ERROR creating HTTP request for '%s': %v",
	smurl, err)
	}
	reqContext, reqCtxCancel := context.WithTimeout(context.Background(), 40*time.Second)
	req = req.WithContext(reqContext)
	rsp, rsperr := b.HSMGlobals.SVCHttpClient.Do(req)

is the particular bit of code for the example in the OP.

We do use the same HTTP client wrapper in several projects, though I didn't review to see if they also pass it a logger without levels.