Make max_blobs_per_block a config parameter #6329

pawanjay176 · 2024-08-29T23:53:01Z

Issue Addressed

N/A

Proposed Changes

Change max_blobs_per_block from a preset value to a config value. This affects a lot of the codebase where we have to use runtime variants of ssz List and Vector.

jimmygchen · 2024-09-05T04:41:19Z

beacon_node/beacon_chain/src/data_availability_checker/overflow_lru_cache.rs

        let (block, blobs_vec) =
            generate_rand_block_and_blobs::<E>(ForkName::Deneb, NumBlobs::Random, &mut rng);
-        let mut blobs: FixedVector<_, <E as EthSpec>::MaxBlobsPerBlock> = FixedVector::default();
+        let max_len = spec.max_blobs_per_block(block.epoch()) as usize;


The epoch here will be random right?

jimmygchen · 2024-09-05T04:48:51Z

consensus/types/src/runtime_var_list.rs

@@ -214,6 +262,85 @@ where
    }
 }

+/// Emulates a SSZ `Vector`.
+#[derive(Clone, Debug)]
+pub struct RuntimeFixedList<T> {


oh we should probably move this to a separate file

jimmygchen · 2024-09-05T04:55:02Z

beacon_node/execution_layer/src/test_utils/execution_block_generator.rs

@@ -676,7 +676,8 @@ impl<E: EthSpec> ExecutionBlockGenerator<E> {
            ForkName::Deneb | ForkName::Electra => {
                // get random number between 0 and Max Blobs
                let mut rng = self.rng.lock();
-                let num_blobs = rng.gen::<usize>() % (E::max_blobs_per_block() + 1);
+                // TODO(pawan): thread the chainspec value here somehow
+                let num_blobs = rng.gen::<usize>() % 6;


Suggested change

let num_blobs = rng.gen::<usize>() % 6;

let num_blobs = rng.gen::<usize>() % (6 + 1);

jimmygchen · 2024-10-09T12:17:30Z

consensus/types/src/chain_spec.rs

+        if self.is_peer_das_enabled_for_epoch(epoch) {
+            self.max_blobs_per_block
+        } else {
+            default_max_blobs_per_block()


Is this necessary? If we're changing this on devnets, then all clients would be using MAX_BLOBS_PER_BLOCK from deneb, and we won't be able to interop with other clients. This config value is already configurable for some clients for Deneb, so if we're overriding this value, other clients would be using this value even before PeerDAS.

For PeerDAS purpose I guess we'd be using a new config value when the time comes?

I think I see what you're trying to do here - this function returns the max_blobs_per_block for the epoch, so in the future it could return MAX_BLOBS_PER_BLOCK for deneb, and MAX_BLOBS_PER_BLOCK_ELECTRA if we're in Electra?

The issue i see with returning the default is that if we start a devnet on Deneb with a different MAX_BLOBS_PER_BLOCK, we won't be able to interop with other clients (which is the main purpose of making this field configurable)

jimmygchen · 2024-10-09T12:34:40Z

beacon_node/beacon_chain/src/data_availability_checker/overflow_lru_cache.rs

        Self {
            block_root,
-            verified_blobs: FixedVector::default(),
+            // TODO(pawan): just make this a vec potentially


I think we can - this was previous an SSZ type because of overflow LRU cache and persisting to disk, but that's been dropped.

jimmygchen · 2024-10-09T12:35:03Z

consensus/types/src/runtime_var_list.rs

@@ -214,6 +262,85 @@ where
    }
 }

+/// Emulates a SSZ `Vector`.


This doesn't really cover SSZ encoding and decoding

jimmygchen · 2024-10-09T12:37:29Z

I've just added the 6.0.0 label to this - I think it would be great if we can release this with the getBlobsV1 changes, as we're keen to get some data on blob count increase with the optimisation. Ideally this get merged relatively soon given the amount of files touched, however this is a nice to have, so we can drop this in favour of higher priority issues.

jimmygchen · 2024-10-09T12:49:09Z

consensus/types/src/runtime_var_list.rs

+    }
+
+    pub fn set_max_len(&mut self, max_len: usize) {
+        self.max_len = Some(max_len);


probably should try to catch unexpected mutations after the max_len is already set, or only allow it to be set once (would OnceCell be an overkill for this?)

Actually this function is not used at all - maybe we could drop it? it feels like it could be a potential footgun

jimmygchen · 2024-10-09T12:54:09Z

consensus/types/src/runtime_var_list.rs

+    ///
+    /// No mutating operation can be performed on an uninitialized instance
+    /// without first setting `max_len`.
+    pub fn empty_uninitialized() -> Self {


feels like this function is for convenience? the little downside is having to make max_len optional when it's known most of the time. But I guess the convenience is worth it if we don't need to mutate it all.

jimmygchen · 2024-10-09T13:10:58Z

beacon_node/http_api/src/block_id.rs

@@ -296,17 +296,24 @@ impl BlockId {
                    ))
                })?
        } else {
-            BlobSidecarList::default()
+            BlobSidecarList::empty_uninitialized()


max_len is not specified here - i think it's not an issue since it's not encoded when BlobSidecarList is used as a standalone object

jimmygchen · 2024-10-09T13:12:53Z

beacon_node/http_api/src/block_id.rs

-                    .map_err(|e| warp_utils::reject::custom_server_error(format!("{:?}", e)))?
+                if let Some(max_len) = list
+                    .first()
+                    .map(|sidecar| chain.spec.max_blobs_per_block(sidecar.epoch()))


we could use the block slot so we could avoid using first?

jimmygchen · 2024-10-09T13:17:58Z

beacon_node/lighthouse_network/src/rpc/methods.rs

    pub fn max_blobs_requested<E: EthSpec>(&self) -> u64 {
-        self.count.saturating_mul(E::max_blobs_per_block() as u64)
+        self.count.saturating_mul(MAX_BLOBS_PER_BLOCK_CEILING)


I think this may cause the node to hit the rate limit quicker because each range blob request will now consume max_blobs_requested tokens from the quota

jimmygchen · 2024-10-09T13:20:15Z

beacon_node/lighthouse_network/src/rpc/methods.rs

+/// bounds checking and other non-consensus critical operations.
+///
+/// For exact value, we should always check the chainspec.
+pub const MAX_BLOBS_PER_BLOCK_CEILING: u64 = 16;


I think we could even go with a higher number if it's for non critical checks and preventing DOS, so we don't run into issues in upcoming upgrades and we don't have to frequently update this?

I thought about this for #6462, but as far as I understood this value may change per fork right?

Yeah it can change, but like i mentioned in the comment, it is to be used only for rate limiting and bounds checking of the incoming rpc request, so we can bump the number up in preparation for a fork

jimmygchen · 2024-10-09T13:36:53Z

beacon_node/network/src/sync/network_context.rs

+                        let max_len = if let Some(blob) = blobs.first() {
+                            self.chain.spec.max_blobs_per_block(blob.epoch()) as usize
+                        } else {
+                            6


add_response only returns Some if all blob sidecars have been received, I think this may be unreachable? if we reach here it means add_response is buggy and might be worth doing some error handling?

jimmygchen · 2024-10-09T13:44:17Z

beacon_node/store/src/hot_cold_store.rs

+                    BlobSidecarList::from_vec(blobs, max_blobs_per_block as usize)
+                } else {
+                    // This always implies that there were no blobs for this block_root
+                    BlobSidecarList::empty_uninitialized()


This is fine for now I think, but note it may break if we end up encoding this within another object like CacheItem or AvailableBlock (I don't see any use of encoding these rn)

jimmygchen · 2024-10-09T13:49:30Z

We also need to move MAX_BLOBS_PER_BLOCK from presets to config YAML files.

jimmygchen

I've added some comments, let me know what you think, thanks!

jxs

Hi Pawan, this is awesome, I wasn't aware of this PR but this is what I had in mind with #6462, left a comment

jxs · 2024-10-21T14:26:36Z

beacon_node/lighthouse_network/src/rpc/methods.rs

+/// bounds checking and other non-consensus critical operations.
+///
+/// For exact value, we should always check the chainspec.
+pub const MAX_BLOBS_PER_BLOCK_CEILING: u64 = 16;


I thought about this for #6462, but as far as I understood this value may change per fork right?

pawanjay176 added 7 commits August 29, 2024 16:11

First pass

25feedf

Add restrictions to RuntimeVariableList api

4dc6e65

Use empty_uninitialized and fix warnings

a9cb329

Fix some todos

60100fc

Merge branch 'unstable' into max-blobs-preset

13f9bba

Fix take impl on RuntimeFixedList

e71020e

cleanup

52bb581

pawanjay176 force-pushed the max-blobs-preset branch from 8b665b5 to 52bb581 Compare September 4, 2024 02:13

pawanjay176 added 2 commits September 4, 2024 12:47

Fix test compilations

d37733b

Fix some more tests

12c6ef1

pawanjay176 marked this pull request as ready for review September 4, 2024 23:19

jimmygchen reviewed Sep 5, 2024

View reviewed changes

Fix test from unstable

2fcb293

jimmygchen added the ready-for-review The code is ready for review label Oct 4, 2024

jimmygchen self-assigned this Oct 4, 2024

jimmygchen reviewed Oct 9, 2024

View reviewed changes

jimmygchen added the v6.0.0 New major release for hierarchical state diffs label Oct 9, 2024

jimmygchen added under-review A reviewer has only partially completed a review. and removed ready-for-review The code is ready for review labels Oct 9, 2024

jimmygchen reviewed Oct 9, 2024

View reviewed changes

jimmygchen requested changes Oct 9, 2024

View reviewed changes

jimmygchen added waiting-on-author The reviewer has suggested changes and awaits thier implementation. and removed under-review A reviewer has only partially completed a review. labels Oct 9, 2024

jimmygchen mentioned this pull request Oct 21, 2024

Fix BlobsByRange by reverting PR6462 #6526

Merged

jxs reviewed Oct 21, 2024

View reviewed changes

Merge branch 'unstable' into max-blobs-preset

21ecb58

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Make max_blobs_per_block a config parameter #6329

Make max_blobs_per_block a config parameter #6329

pawanjay176 commented Aug 29, 2024

jimmygchen Sep 5, 2024

jimmygchen Sep 5, 2024

jimmygchen Sep 5, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen commented Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jxs Oct 21, 2024

pawanjay176 Oct 21, 2024

jimmygchen Oct 9, 2024

jimmygchen Oct 9, 2024

jimmygchen commented Oct 9, 2024

jimmygchen left a comment

jxs left a comment

jxs Oct 21, 2024

	let num_blobs = rng.gen::<usize>() % 6;
	let num_blobs = rng.gen::<usize>() % (6 + 1);

Make max_blobs_per_block a config parameter #6329

Are you sure you want to change the base?

Make max_blobs_per_block a config parameter #6329

Conversation

pawanjay176 commented Aug 29, 2024

Issue Addressed

Proposed Changes

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jimmygchen commented Oct 9, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jimmygchen commented Oct 9, 2024

jimmygchen left a comment

Choose a reason for hiding this comment

jxs left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment