Integrate a new custom ipld module into laconicd #24

Open
opened 2022-08-31 16:40:56 +00:00 by i-norden · 4 comments
Member

From Ashwin: We can build this as a library instead of a module (as it doesn't contribute to state)

  • CID storage/retrieval functionality exposed as a cosmos-sdk module called ipfs
  • Connects to postgres, saves/loads data from public.blocks table
  • Registry and other cosmos-sdk modules should be able to load data from ipfs
From Ashwin: We can build this as a library instead of a module (as it doesn't contribute to state) * CID storage/retrieval functionality exposed as a cosmos-sdk module called `ipfs` * Connects to postgres, saves/loads data from `public.blocks` table * Registry and other cosmos-sdk modules should be able to load data from ipfs
Owner

Bugfixes then this feature please, it should be a light lift.

Bugfixes then this feature please, it should be a light lift.
Author
Member

@0xmuralik @aleem1314

For some background:

As a Tendermint/CosmosSDK state machine, some of the main roles of Laconicd are

  1. Serve as a registry for watchers, service providers, bond associations, etc
  2. Facilitate governance decisions e.g. trigger protocol upgrades and trigger service provider seat auctions
  3. Facilitate watcher service auctions whereby service providers bid on watcher service contracts
  4. On-chain watcher fraud proof evaluation

Watchers can be thought of as WASM bytes that describe a custom state transition function over custom subsets of native (e.g. Ethereum) chain state.

To support this arbitrary transforming and indexing of chain data (while also preserving its Merkle context), we first extract chain state from its "flat" Merkle representation on disc (e.g. levelDB for geth) into an initial relational representation defined here. This is done by a modified geth client and some other tools for historical state: https://github.com/cerc-io/ipld-eth-state-snapshot, https://github.com/cerc-io/eth-statediff-service. We refer to this whole stack as the "full-indexing node" or FIN.

The ipld-eth-db representation retains the native Merkle representation (aka hash => node mappings for every node in the entire ethereum DAG) but in an IPFS blockstore format. An IPFS blockstore (v1) is a flat key-value store that maps multihash keys to IPLD blocks (in the case of most Ethereum objects, these are RLP encoded bytes), the shim for treating a Postgres database as a v1 blockstore is https://github.com/ipfs/go-ds-sql, so we currently store our IPLD blocks in that format (except note that we have denormalized our table by block_number). Additionally the ipld-eth-db schema has other tables (e.g. "eth.header_cids", "eth.state_cids") which "blowout" certain ethereum object fields into searchable columns and reference back to the raw IPLD record.

I mention v1 blockstore because that is what we currently use and is what go-ds-sql and ipfs work with, but we are in the middle of upgrading ipld-eth-db to v5 and one of the main changes is that we revert to using a v0 blockstore format. That is, instead of using multihash keys we use fully fledged CIDs. Some additional context on that change: https://github.com/cerc-io/ipld-eth-db/issues/122.

The reason I am providing all this context here is because up until this issue laconicd has been developed in isolation from the FIN stack, but this issue is where the two finally begin to converge.

In order for laconic to perform watcher fraud proof evaluation (4 above), in order to evaluate a fraud proof from within the laconic Tendermint/Cosmos state machine, we need to be able to provide the state machine access to the raw chain inputs to the watcher (the fraud proof evaluation essentially involves repeating the challenged state transitions in the federated laconicd execution environment to check and see if you get the expected result or not).

That is, we need a module that exposes a keeper with a basic CRUD interface on-top of ipld-eth-db (specifically, the “ipld.blocks” table) instead of on top of the Cosmos Multistore/KVStore/IAVL.

Table we need to interface with:

CREATE TABLE IF NOT EXISTS ipld.blocks (
    block_number BIGINT NOT NULL,
    key TEXT NOT NULL,
    data BYTEA NOT NULL,
    PRIMARY KEY (key, block_number)
);

key is a CID, and data is an IPLD block bytes.

To summarize, we need to write a module that exposes a interface that has
“Get”, “Set”, “Delete”, “Update” operations for the ipld.blocks table of ipld-eth-db. We can extend/refine this interface further as we begin to use it.

I am going to try to break this down into some finer actionable tasks today.

@0xmuralik @aleem1314 For some background: As a Tendermint/CosmosSDK state machine, some of the main roles of Laconicd are 1. Serve as a registry for watchers, service providers, bond associations, etc 2. Facilitate governance decisions e.g. trigger protocol upgrades and trigger service provider seat auctions 3. Facilitate watcher service auctions whereby service providers bid on watcher service contracts 4. On-chain watcher fraud proof evaluation Watchers can be thought of as WASM bytes that describe a custom state transition function over custom subsets of native (e.g. Ethereum) chain state. To support this arbitrary transforming and indexing of chain data (while also preserving its Merkle context), we first extract chain state from its "flat" Merkle representation on disc (e.g. levelDB for geth) into an initial relational representation defined [here](https://github.com/cerc-io/ipld-eth-db). This is done by a [modified geth client](https://github.com/cerc-io/go-ethereum/tree/v1.10.26-statediff-v4/statediff) and some other tools for historical state: https://github.com/cerc-io/ipld-eth-state-snapshot, https://github.com/cerc-io/eth-statediff-service. We refer to this whole stack as the "full-indexing node" or FIN. The ipld-eth-db representation retains the native Merkle representation (aka hash => node mappings for every node in the entire [ethereum DAG](https://github.com/ipld/ipld/tree/master/specs/codecs/dag-eth)) but in an IPFS blockstore format. An IPFS blockstore (v1) is a flat key-value store that maps multihash keys to IPLD blocks (in the case of most Ethereum objects, these are RLP encoded bytes), the shim for treating a Postgres database as a v1 blockstore is https://github.com/ipfs/go-ds-sql, so we currently store our IPLD blocks in that format (except note that we have denormalized our table by block_number). Additionally the ipld-eth-db schema has other tables (e.g. "eth.header_cids", "eth.state_cids") which "blowout" certain ethereum object fields into searchable columns and reference back to the raw IPLD record. I mention v1 blockstore because that is what we currently use and is what go-ds-sql and ipfs work with, but we are in the middle of upgrading ipld-eth-db to v5 and one of the main changes is that we revert to using a v0 blockstore format. That is, instead of using multihash keys we use fully fledged CIDs. Some additional context on that change: https://github.com/cerc-io/ipld-eth-db/issues/122. The reason I am providing all this context here is because up until this issue laconicd has been developed in isolation from the FIN stack, but this issue is where the two finally begin to converge. In order for laconic to perform watcher fraud proof evaluation (4 above), in order to evaluate a fraud proof from within the laconic Tendermint/Cosmos state machine, we need to be able to provide the state machine access to the raw chain inputs to the watcher (the fraud proof evaluation essentially involves repeating the challenged state transitions in the federated laconicd execution environment to check and see if you get the expected result or not). That is, we need a module that exposes a <s>keeper with a</s> basic CRUD interface on-top of ipld-eth-db (specifically, the “ipld.blocks” table) instead of on top of the Cosmos Multistore/KVStore/IAVL. Table we need to interface with: ```sql CREATE TABLE IF NOT EXISTS ipld.blocks ( block_number BIGINT NOT NULL, key TEXT NOT NULL, data BYTEA NOT NULL, PRIMARY KEY (key, block_number) ); ``` `key` is a CID, and `data` is an IPLD block bytes. To summarize, we need to write a module that exposes a interface that has “Get”, “Set”, “Delete”, “Update” operations for the [ipld.blocks](https://github.com/cerc-io/ipld-eth-db/blob/v5/db/migrations/00001_create_ipfs_blocks_table.sql) table of ipld-eth-db. We can extend/refine this interface further as we begin to use it. I am going to try to break this down into some finer actionable tasks today.
Author
Member

This won't need to be a x/module with a Keeper, as it isn't managing state committed to Tendermint, it can just be some package that other modules/keepers pull in.

I think we could actually get away with using https://github.com/cerc-io/ipfs-ethdb for now as it exposes the basic CRUD operations we need on-top of ipld-eth-db (to start experimenting with anyways), but it brings along some unnecessary stuff.

This won't need to be a x/module with a Keeper, as it isn't managing state committed to Tendermint, it can just be some package that other modules/keepers pull in. I think we could actually get away with using https://github.com/cerc-io/ipfs-ethdb for now as it exposes the basic CRUD operations we need on-top of ipld-eth-db (to start experimenting with anyways), but it brings along some unnecessary stuff.
Author
Member

Steps to close this issue:

  1. Write simple library that exposes a CRUD interface on top of our Postgres IPLD blockstore. This can be heavily inspired by ipfs-ethdb, taking the work from there and trimming away everything but Get, Set, Has, Delete.
  2. Finish up record content type work
  3. Finish up WASM work

We could begin work on 1 before finishing the other lines of work, but those are the two places in laconicd where we will need to provide access to the IPLD blockstore so we won't be able to integrate the package further until that work is done.

Steps to close this issue: 1. Write simple library that exposes a CRUD interface on top of our Postgres IPLD blockstore. This can be heavily inspired by [ipfs-ethdb](https://github.com/cerc-io/ipfs-ethdb/blob/v5/postgres/database.go#L128), taking the work from there and trimming away everything but Get, Set, Has, Delete. 2. Finish up record content type work 3. Finish up WASM work We could begin work on 1 before finishing the other lines of work, but those are the two places in laconicd where we will need to provide access to the IPLD blockstore so we won't be able to integrate the package further until that work is done.
Sign in to join this conversation.
No Milestone
No project
No Assignees
2 Participants
Notifications
Due Date
The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: cerc-io/laconicd-deprecated#24
No description provided.