Home

Awesome

Index Provider :loudspeaker:

Go Reference Coverage Status

A golang implementation of index provider

This repo provides a reference index provider implementation that can be used to advertise content to indexer nodes and serve retrieval requests over graphsync both as a standalone service or embedded into an existing Golang application via a reusable library.

A list of features include:

Use of the ipni-cli provides additional utility that is useful to check the functioning of an index-provider instance:

Current status :construction:

This implementation is under active development.

Background

The protocol implemented by this repository is the index provider portion of a larger indexing protocol documented here . The indexer node implementation can be found at storetheindex and go-libipni.

For more details on the ingestion protocol itself see IPNI Spec - Ingestion .

Install

Prerequisite:

To use the provider as a Go library, execute:

go get github.com/ipni/index-provider

To install the latest provider CLI, run:

go install github.com/ipni/index-provider/cmd/provider@latest

Usage

Running an standalone provider daemon

To run a provider service first initialize it by executing:

provider init

Initialization generates a default configuration for the provider instance along with a randomly generated identity keypair. The configuration is stored at user home under .index-provider/config in JSON format. The root configuration path can be overridden by setting the PROVIDER_PATH environment variable

Once initialized, start the service daemon by executing:

provider daemon

The running daemon allows advertisement for new content to the indexer nodes and retrieval of content over GraphSync. Additionally, it starts an admin HTTP server that enables administrative operations using the provider CLI tool. By default, the admin server is bound to http://localhost:3102.

You can then advertise content by importing/removing CAR files via the provider CLI, for example:

provider import car -i <path-to-car-file>

Both CARv1 and CARv2 formats are supported. Index is regenerated on the fly if one is not present.

Exposing delegated routing server from provider (Experimental)

Provider can export a Delegated Routing server. Delegated Routing allows IPFS nodes to advertise their contents to indexers alongside DHT. Delegated Routing server is off by default. To enable it, add the following configuration block to the provider config file.

{
  ...
  DelegatedRouting {
    ListenMultiaddr: "/ip4/0.0.0.0/tcp/50617 (example)"
  }
  ...
}

Configuring Kubo to advertise content onto IPNI (Experimental)

Disclaimer: PUT /routing/v1 is currently not officially supported in Kubo. Please use it at your own risk. See IPIP-378 for the latest updates.

Kubo supports HTTP delegated routing as of v0.18.0. The following section contains configuration examples and a few tips to enable Kubo to advertise its CIDs to IPNI systems like cid.contact using index-provider. Delegated Routing is still in the Experimental stage and configuration might change from version to version. This section serves as an inspiration for configuring your node to use IPNI, but for comprehensive information, refer to the Kubo documentation. Here are some important points to consider:

To configure index-provider to expose the delegated routing server, use the following configuration:

"DelegatedRouting": {
  "ListenMultiaddr": "/ip4/0.0.0.0/tcp/50617",
  "ProviderID": "PEER ID OF YOUR IPFS NODE",
  "Addrs": [] // List of multiaddresses that you'd like to be advertised to IPNI. Announce addrs are going to be advertised if not specified.
}

Configure Kubo to publish into both DHT and IPNI:

"Routing": {
    "Methods": {
      "find-peers": {
        "RouterName": "WanDHT"
      },
      "find-providers": {
        "RouterName": "ParallelHelper"
      },
      "get-ipns": {
        "RouterName": "WanDHT"
      },
      "provide": {
        "RouterName": "ParallelHelper"
      },
      "put-ipns": {
        "RouterName": "WanDHT"
      }
    },
    "Routers": {
      "IndexProvider": {
        "Parameters": {
          "Endpoint": "http://127.0.0.1:50617",
          "MaxProvideBatchSize": 10000,
          "MaxProvideConcurrency": 1
        },
        "Type": "http"
      },
      "ParallelHelper": {
        "Parameters": {
          "Routers": [
            {
              "IgnoreErrors": true,
              "RouterName": "IndexProvider",
              "Timeout": "30m"
            },
            {
              "IgnoreErrors": true,
              "RouterName": "WanDHT",
              "Timeout": "30m"
            }
          ]
        },
        "Type": "parallel"
      },
      "WanDHT": {
        "Parameters": {
          "AcceleratedDHTClient": false,
          "Mode": "auto",
          "PublicIPNetwork": true
        },
        "Type": "dht"
      }
    },
    "Type": "custom"
  },

With the above configuration, Kubo will advertise its CIDs to both DHT and IPNI and will use both DHT and IPNI for find-providers lookups. Additionally, enable the following flag in the Kubo config to enable batch re-provides (especially for larger nodes):

"Experimental": { 
 "AcceleratedDHTClient": true,
},

After adding a new file to your Kubo node, you should see index-provider logs starting to appear immediately. If that doesn't happen, it's likely that Kubo has been configured incorrectly.

index-provider publishes announcements about new advertisements on a libp2p pub/sub topic. This topic is listened by IPNI systems like cid.contact. Once a new announcement is seen, IPNI would reach out to index-provider to download advertisement chains and index the content. It's important to keep in mind:

Here are a few additional configuration options to consider:

Embedding index provider integration

The root go module offers a set of reusable libraries that can be used to embed index provider support into existing application. The core provider.Interface is implemented by engine.Engine.

The provider Engine exposes a set of APIs that allows a user to programmatically announce the availability or removal of content to the indexer nodes referred to as “advertisement”. Advertisements are represented as an IPLD DAG, chained together via a link to the previous advertisement. An advertisement effectively captures the "diff" of the content that is either added or is no longer provided.

Each advertisement contains:

The Entries link points to the IPLD node that contains a list of mulitihashes being advertised. The list is represented as a chain of "Entry Chunk"s where each chunk contains a list of multihashes and a link to the next chunk. This is to accommodate pagination for a large number of multihashes.

The engine can be configured to dynamically look up the list of multihashes that correspond to the context ID of an advertisement. To do this, the engine requires a MultihashLister to be registered. The MultihashLister is then used to look up the list of multihashes associated to a content advertisement.

NotifyPut and NotifyRemove are convinience wrappers on top of Publish that aim to help to construct advertisements. They take care of such things as generating entry chunks, linking to the last published advertisement, signing and others. NotifyPut can be also used to update metadata for a previously published advertisement (for example in the case when a protocol has changed). That can be done by invoking NotifyPut with the same context ID but different metadata field. ErrAlreadyAdvertised will be returned if both context ID and metadata have stayed the same.

For an example on how to start up a provider engine, register a lister and advertise content, see:

Configuration for Sublishing Advertisements

See the Publisher Configuratgion document

Publishing advertisements with extended providers

Extended providers field allows for specification of provider families, in cases where a provider operates multiple PeerIDs, perhaps with different transport protocols between them, but over the same database of content.

ExtendedProviders can either be applied at the chain-level (for all previous and future CIDs published by a provider) or at a context-level (for CIDs belonging to the specified context ID). That behaviour is set by ContextID field. Multiple different behaviours can be triggered by a combination of ContextID, Metadata, ExtendedProviders and Override fields. For more information see the specification

Advertisements with ExtendedProviders can be composed manually or by using a convenience ExtendedProvidersAdBuilder and will have to be signed by the main provider as well as by all ExtendedProviders' identities. Private keys for these identities have to be provided in the xproviders.Info (objects)[https://github.com/ipni/index-provider/blob/main/engine/xproviders/xproviders.go] and
ExtendedProvidersAdBuilder will take care of the rest.

Identity of the main provider will be added to the extended providers list automatically and should not be passed in explicitly.

Some examples can be found below (assumes the readers familiriaty with the (specification)[https://github.com/ipni/storetheindex/blob/main/doc/ingest.md#extendedprovider]).

Publishing an advertisement with context-level ExtendedProviders, that will be returned only for CIDs from the specified context ID:

  adv, err := ep.NewExtendedProviderAdBuilder(providerID, priv, addrs).
    WithContextID(contextID). 
    WithMetadata(metadata). 
    WithOverride(override). 
    WithExtendedProviders(extendedProviders). 
    WithLastAdID(lastAdId). 
    BuildAndSign()

  if err != nil {
    //...
  }

  engine.Publish(ctx, *adv)
)

Constructing an advertisement with chain-level ExtendedProviders, that will be returned for every past and future CID published by the main provider:

  adv, err := ep.NewExtendedProviderAdBuilder(providerID, priv, addrs). 
    WithMetadata(metadata). 
    WithExtendedProviders(extendedProviders).
    WithLastAdID(lastAdId). 
    BuildAndSign()
)

ExtendedProviders can also be used to add a new metadata to all CIDs published by the main provider. Such advertisement need to be constructed without specifying ExtendedProviders at all. The identity of the main provider will be added to the ExtendedProviders list by the builder automatically, so the resulting advertisement will contain only the main provider in the ExtendedProviders list (yes, this is also possible:). That can be used for example to advertise new endpoint with a new protocol alongside the previously advertised one:

  adv, err := ep.NewExtendedProviderAdBuilder(providerID, priv, addrs). 
    WithMetadata(metadata). 
    WithLastAdID(lastAdId). 
    BuildAndSign()
)

On ingestion, previously published ExtendedProviders get overwritten (not merged!) by the newer ones. So in order to update ExtendedProviders, just publish a new chain/context-level advertisement with the required changes.

Examples of constructing advertisements with ExtendedProviders can be found (here)[https://github.com/ipni/index-provider/blob/main/engine/xproviders/xproviders_test.go].

provider CLI

The provider CLI can be used to interact with a running daemon via the admin server to perform a range of administrative operations. For example, the provider CLI can be used to import a CAR file and advertise its content to the indexer nodes by executing:

provider import car -i <path-to-car-file>

For usage description, execute provider --help

Storage Consumption

The index provider engine uses a given datastore to persist two general category of data:

  1. Internal advertisement mappings, and
  2. Chunked entries chain cache

If the datastore passed to the engine is reused, it is recommended to wrap it in a namespace prior to instantiating the engine.

Internal advertisement mappings

The internal advertisement mappings are purely used by the engine to efficiently handle publication requests. It generally includes:

The storage consumed by such mappings is negligible and grows linearly as a factor of the number of advertisements published.

Chunked entries chain cache

This category stores chunked entries generated by publishing an advertisement with a never seen before context ID. The chunks are stored in an LRU cache, the maximum size of which is configured by the following configuration parameters in Ingest config:

The exact storage usage depends on the size of multihashes. For example, using the default config to advertise 128-bit long multihashes will result in chunk sizes of 0.25MiB with maximum cache growth of 256 MiB.

To delete the cache set PurgeLinkCache to true and restart the engine.

Note that the LRU cache may grow beyond its max size if the generated chain of chunks is longer than the configured LinkChunkSize. This is to avoid partial caching of chunks within a single advertisement. The cache expansion is logged in INFO level at provider/engine logging subsystem.

Related Resources

License

SPDX-License-Identifier: Apache-2.0 OR MIT