feat: basic support for Ipfs-Path-Affinity from IPIP-462 #592

lidel · 2024-03-22T23:56:43Z

This PR aims to kick-off discussion how we would support IPIP-462 in boxo/gateway.

For more info and header semantics, see ipfs/specs#462

proposed scope

What I want to do for now is to add minimal code to start leveraging Ipfs-Path-Affinity hints within existing boxo/gateway codebase, so we cna deploy it to our gateways and allow clients like service-worker-gateway or ipfs-chromium to pass hint and retrieve content even when internal CID was not announced directly.

support percent-encoded values
limit to 3 hints per request
cancel work if original request finished
skip work that would not produce any benefit in routing (if path from hint is already cached it won't trigger any additional provider lookups)

future scope

We may formalize this by extending IPFSBackend with explicit place to pass this hint, and then wire it up into routing system of Kubo, but it feels more involved and would like to do that in follow-up, rather than block on it here.

This is first stab at leveraging these hints withing existing boxo/gateway codebase. It is pretty blunt, but will enable smart clients fetching sub-DAGs to work around any content routing gaps For more info and header semantics see ipfs/specs#462

gateway/handler.go

+		// Confirm it is a valid content path
+		affinityPath, err := path.NewPath(headerValue)
+		if err != nil {
+			logger.Debugw("skipping invalid Ipfs-Path-Affinity hint", "error", err)


gateway/handler.go

+		// Skip duplicated work if immutable affinity hint is a subset of requested immutable contentPath
+		// (protect against broken clients that use affinity incorrectly)
+		if !contentPath.Mutable() && !affinityPath.Mutable() && strings.HasPrefix(contentPath.String(), affinityPath.String()) {
+			logger.Debugw("skipping redundant Ipfs-Path-Affinity hint", "affinity", affinityPath)


gateway/handler.go

+		// Process hint in background without blocking response logic for contentPath
+		go func(contentPath path.Path, affinityPath path.Path, logger *zap.SugaredLogger) {
+			var immutableAffinityPath path.ImmutablePath
+			logger.Debugw("async processing of Ipfs-Path-Affinity hint", "affinity", affinityPath)


gateway/handler.go

+			if affinityPath.Mutable() {
+				// Skip work if mutable affinity hint is a subset of mutable contentPath
+				if contentPath.Mutable() && strings.HasPrefix(contentPath.String(), affinityPath.String()) {
+					logger.Debugw("skipping redundant Ipfs-Path-Affinity hint", "affinity", affinityPath)


gateway/handler.go

+				}
+				immutableAffinityPath, _, _, err = i.backend.ResolveMutable(r.Context(), affinityPath)
+				if err != nil {
+					logger.Debugw("error while resolving mutable Ipfs-Path-Affinity hint", "affinity", affinityPath, "error", err)


gateway/handler.go

+				}
+				immutableAffinityPath, _, _, err = i.backend.ResolveMutable(r.Context(), affinityPath)
+				if err != nil {
+					logger.Debugw("error while resolving mutable Ipfs-Path-Affinity hint", "affinity", affinityPath, "error", err)


gateway/handler.go

+				// original contentPath was received and returned to HTTP
+				// client before below get is done, the work is cancelled.
+
+				logger.Debugw("started async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath)


gateway/handler.go

+
+				logger.Debugw("started async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath)
+				_, _, err = i.backend.GetBlock(r.Context(), immutableAffinityPath)
+				logger.Debugw("ended async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath, "error", err)


gateway/handler.go

+
+				logger.Debugw("started async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath)
+				_, _, err = i.backend.GetBlock(r.Context(), immutableAffinityPath)
+				logger.Debugw("ended async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath, "error", err)


gateway/handler.go

+				_, _, err = i.backend.GetBlock(r.Context(), immutableAffinityPath)
+				logger.Debugw("ended async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath, "error", err)
+			} else {
+				logger.Debugw("skipping Ipfs-Path-Affinity hint due to data being locally cached", "affinity", affinityPath)


codecov · 2024-03-23T00:02:16Z

Codecov Report

Attention: Patch coverage is 11.11111% with 64 lines in your changes are missing coverage. Please review.

Project coverage is 59.69%. Comparing base (b101ba0) to head (39a8bd7).

@@            Coverage Diff             @@
##             main     #592      +/-   ##
==========================================
- Coverage   59.78%   59.69%   -0.10%     
==========================================
  Files         232      232              
  Lines       28124    28197      +73     
==========================================
+ Hits        16815    16833      +18     
- Misses       9837     9892      +55     
  Partials     1472     1472

Files	Coverage Δ
gateway/handler.go	`70.63% <11.11%> (-7.65%)`	⬇️

... and 15 files with indirect coverage changes

hacdias · 2024-03-25T12:18:42Z

gateway/handler.go

@@ -233,6 +233,8 @@ func (i *handler) getOrHeadHandler(w http.ResponseWriter, r *http.Request) {
 		return
 	}

+	i.handlePathAffinityHints(w, r, contentPath, logger)


What about only calling this if it's a request for raw blocks or CAR files as indicated here? Or do you want to extend the scope?

hacdias

It seems like a good initial version, as it will trigger the search for providers for the affinity path, which are likely the same providers that contain the direct block.

I pushed a quick fix for the gateway sharness tests.

SgtPooki · 2024-04-11T10:44:36Z

gateway/handler.go

@@ -233,6 +233,8 @@ func (i *handler) getOrHeadHandler(w http.ResponseWriter, r *http.Request) {
 		return
 	}

+	i.handlePathAffinityHints(w, r, contentPath, logger)
+


As someone not familiar with go nor this codebase, these variable names need work 😳

@SgtPooki In Go HTTP server request handlers, w and r are pretty much convention for http.ResponseWriter and *http.Request respectively. This is also seen the Go docs and examples.

SgtPooki · 2024-04-11T10:48:42Z

gateway/handler.go

+			logger.Debugw("async processing of Ipfs-Path-Affinity hint", "affinity", affinityPath)
+			if affinityPath.Mutable() {
+				// Skip work if mutable affinity hint is a subset of mutable contentPath
+				if contentPath.Mutable() && strings.HasPrefix(contentPath.String(), affinityPath.String()) {


Seems like hasPrefix check could be pulled out into a more semantic function since it's used at least twice.

gammazero · 2024-06-23T23:36:36Z

gateway/handler.go

+// These optional hints are mostly useful for trustless block requests.
+// See https://github.com/ipfs/specs/pull/462
+func (i *handler) handlePathAffinityHints(w http.ResponseWriter, r *http.Request, contentPath path.Path, logger *zap.SugaredLogger) {
+	headerName := "Ipfs-Path-Affinity"


Suggested change

headerName := "Ipfs-Path-Affinity"

const headerName = "Ipfs-Path-Affinity"

gammazero · 2024-06-23T23:56:43Z

gateway/handler.go

+			if !i.backend.IsCached(r.Context(), immutableAffinityPath) {
+				// The intention of below code is to asynchronously preconnect
+				// gateway with providers of the affinityPath in
+				// Ipfs-Path-Affinity hint. Once connected, these peers can be
+				// asked directly (via mechanism like bitswap) for blocks
+				// related to main request for contentPath, and retrieve them,
+				// even when no other routing system had them announced. If
+				// original contentPath was received and returned to HTTP
+				// client before below get is done, the work is cancelled.
+
+				logger.Debugw("started async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath)
+				_, _, err = i.backend.GetBlock(r.Context(), immutableAffinityPath)
+				logger.Debugw("ended async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath, "error", err)
+			} else {
+				logger.Debugw("skipping Ipfs-Path-Affinity hint due to data being locally cached", "affinity", affinityPath)
+			}


minor: Instead of if .. else, check if cached and return if it is.

Suggested change

if !i.backend.IsCached(r.Context(), immutableAffinityPath) {

// The intention of below code is to asynchronously preconnect

// gateway with providers of the affinityPath in

// Ipfs-Path-Affinity hint. Once connected, these peers can be

// asked directly (via mechanism like bitswap) for blocks

// related to main request for contentPath, and retrieve them,

// even when no other routing system had them announced. If

// original contentPath was received and returned to HTTP

// client before below get is done, the work is cancelled.

logger.Debugw("started async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath)

_, _, err = i.backend.GetBlock(r.Context(), immutableAffinityPath)

logger.Debugw("ended async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath, "error", err)

} else {

logger.Debugw("skipping Ipfs-Path-Affinity hint due to data being locally cached", "affinity", affinityPath)

}

if i.backend.IsCached(r.Context(), immutableAffinityPath) {

logger.Debugw("skipping Ipfs-Path-Affinity hint due to data being locally cached", "affinity", affinityPath)

return

}

// The intention of below code is to asynchronously preconnect

// gateway with providers of the affinityPath in

// Ipfs-Path-Affinity hint. Once connected, these peers can be

// asked directly (via mechanism like bitswap) for blocks

// related to main request for contentPath, and retrieve them,

// even when no other routing system had them announced. If

// original contentPath was received and returned to HTTP

// client before below get is done, the work is cancelled.

logger.Debugw("started async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath)

_, _, err = i.backend.GetBlock(r.Context(), immutableAffinityPath)

logger.Debugw("ended async search for providers of Ipfs-Path-Affinity hint", "affinity", affinityPath, "error", err)

gammazero · 2024-06-24T00:09:55Z

gateway/handler.go

+		// Skip duplicated work if immutable affinity hint is a subset of requested immutable contentPath
+		// (protect against broken clients that use affinity incorrectly)
+		if !contentPath.Mutable() && !affinityPath.Mutable() && strings.HasPrefix(contentPath.String(), affinityPath.String()) {


The comment is a little confusing about which is a prefix of the other.

if immutable affinity hint is a subset of requested immutable contentPath

It could mean that the contentPath should be the prefix, since the subset is a more specific (longer) path:

contentPath = "/top-level" affinityPath = "/top-level/sub-path"

or it could me that the affinity path is a string subset of the longer content path:

contentPath = "/top-level/stuff/sub-stuff" affinityPath = "/top-level/stuff"

lidel added 2 commits March 22, 2024 19:24

chore: changelog

6d16b69

lidel requested a review from aschmahmann March 22, 2024 23:56

lidel requested review from hacdias and a team as code owners March 22, 2024 23:56

lidel mentioned this pull request Mar 23, 2024

IPIP-462: Ipfs-Path-Affinity on Gateways ipfs/specs#462

Open

github-advanced-security bot found potential problems Mar 23, 2024

View reviewed changes

ci: update to go 1.22 on gateway sharness tests

39a8bd7

hacdias reviewed Mar 25, 2024

View reviewed changes

SgtPooki reviewed Apr 11, 2024

View reviewed changes

gammazero reviewed Jun 24, 2024

View reviewed changes

gammazero marked this pull request as draft October 29, 2024 17:51

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: basic support for Ipfs-Path-Affinity from IPIP-462 #592

feat: basic support for Ipfs-Path-Affinity from IPIP-462 #592

lidel commented Mar 22, 2024 •

edited

Loading

codecov bot commented Mar 23, 2024 •

edited

Loading

hacdias Mar 25, 2024

hacdias left a comment

SgtPooki Apr 11, 2024

gammazero Jul 9, 2024

SgtPooki Apr 11, 2024

gammazero Jun 23, 2024

gammazero Jun 23, 2024

gammazero Jun 24, 2024 •

edited

Loading

	headerName := "Ipfs-Path-Affinity"
	const headerName = "Ipfs-Path-Affinity"

feat: basic support for Ipfs-Path-Affinity from IPIP-462 #592

Are you sure you want to change the base?

feat: basic support for Ipfs-Path-Affinity from IPIP-462 #592

Conversation

lidel commented Mar 22, 2024 • edited Loading

proposed scope

future scope

codecov bot commented Mar 23, 2024 • edited Loading

Codecov Report

hacdias Mar 25, 2024

Choose a reason for hiding this comment

hacdias left a comment

Choose a reason for hiding this comment

SgtPooki Apr 11, 2024

Choose a reason for hiding this comment

gammazero Jul 9, 2024

Choose a reason for hiding this comment

SgtPooki Apr 11, 2024

Choose a reason for hiding this comment

gammazero Jun 23, 2024

Choose a reason for hiding this comment

gammazero Jun 23, 2024

Choose a reason for hiding this comment

gammazero Jun 24, 2024 • edited Loading

Choose a reason for hiding this comment

lidel commented Mar 22, 2024 •

edited

Loading

codecov bot commented Mar 23, 2024 •

edited

Loading

gammazero Jun 24, 2024 •

edited

Loading