PBS Snapshot is a command-line program written in Python that delivers a simplistic snapshot of proposer/builder separation on Ethereum. It runs out of the box with no authentication requirements, however for larger historic snapshots we recommend setting a custom JSON-RPC endpoint and rate limit.
The snapshot is created by:
- Querying all relays to get recently delivered MEV-Boost payloads
- Leveraging cryo to download block data that we cross-reference against the relay payload data
- Calculating the share of blocks proposed using MEV-Boost and counting blocks delivered by each relay to watch for abnormalities
- Downloading missed slot data from the beaconcha.in API to retrieve proposer indexes for missed slots
- Identifying tagged proposers who missed their slots to isolate proposer sets who may be experiencing outages (sourced from Dune Analytics and curated by Hildobby)
PBS Snapshot requires Git and Python 3.12.
$ git clone --depth 1 https://github.com/dataalways/pbs-snapshot
$ cd pbs-snapshotWe strongly recommend installing the dependencies to a virtual environment.
For users of pipenv, we provide both Pipfile and Pipfile.lock. To guarantee reproducible execution, install the dependencies like so:
$ pipenv install --ignore-pipfileWe also provide a minimal pyproject.toml file for users of modern project managers such as Rye, Hatch and PDM.
In the following examples, each command is run in a virtual environment containing the project's dependencies.
(pbs-snapshot) $ ./pbs_snapshot.pypbs_snapshot: INFO: downloading payload delivery data from MEV-Boost relays
pbs_snapshot: INFO: downloading block data from RPC endpoint
pbs_snapshot: INFO: writing data to 'export/slots_9275070-9275119_snapshot.csv'
pbs_snapshot: INFO: downloading missed slot data from beaconcha.in
All queried relays were available!
Range of blocks analyzed: from 20069909 to 20069958
Number of payloads delivered by relay: ultrasound: 30, bloxroute_max_profit: 26, bloxroute_regulated: 17, agnostic: 17, flashbots: 10, aestus: 6
Number of payloads delivered by only one relay: bloxroute: 10, ultrasound: 10, flashbots: 3
Share of MEV-Boost blocks: 96.00%
Share of missed slots: 0.00%
Number of proposers that missed their slots: 0
Number of untagged proposers that missed their slots: 0
Analysis of output:
- There were 0 missed slots in the 10 minutes prior to program execution
- 96% of blocks were built using MEV-Boost, which is in line with expectations
- Network health was nominal
Example 2: bloXroute BDN issues on March 27, 2024
(pbs-snapshot) $ ./pbs_snapshot.py --max-slot 8729000 --lookback 1000pbs_snapshot: INFO: downloading payload delivery data from MEV-Boost relays
pbs_snapshot: INFO: downloading block data from RPC endpoint
pbs_snapshot: INFO: writing data to 'export/slots_8728001-8729000_snapshot.csv'
pbs_snapshot: INFO: writing data to 'export/slots_8728135-8728990_missed.csv'
pbs_snapshot: INFO: downloading missed slot data from beaconcha.in
All queried relays were available!
Range of blocks analyzed: from 19527062 to 19527984
Number of payloads delivered by relay: ultrasound: 564, agnostic: 330, bloxroute_max_profit: 275, bloxroute_regulated: 211, flashbots: 189, aestus: 97, titan: 8, manifold: 3, eden: 2
Number of payloads delivered by only one relay: ultrasound: 200, bloxroute: 105, flashbots: 65, agnostic: 13, eden: 2, aestus: 1, manifold: 1, titan: 1
Share of MEV-Boost blocks: 85.81%
Share of missed slots: 7.70%
Number of missed MEV-Boost payloads delivered by relay: bloxroute: 59, ultrasound: 18, flashbots: 10, agnostic: 9, aestus: 7
Number of proposers that missed their slots: 77
Number of untagged proposers that missed their slots: 16
Number of missed slots by entity: Lido: 22, Figment: 11, Coinbase: 9, Binance: 6, Kiln: 5, CoinSpot: 2, ether.fi: 2, Rocket Pool: 1, Kraken: 1, OKX: 1, Staked.us: 1
Analysis of output:
- Data was available for all relays
- There were 77 missed slots; bloXroute relayed the payloads for 59 (77%) of those slots
- The proposers were a diverse set with no group significantly over-represented versus their market share
- The most likely cause of the network instability was an issue with bloXroute's relays
Example 3: Coinbase Outage on March 8, 2024
(pbs-snapshot) $ ./pbs_snapshot.py --max-slot 8593000 --lookback 500pbs_snapshot: INFO: downloading payload delivery data from MEV-Boost relays
pbs_snapshot: INFO: relay 'bloxroute_max_profit' returned no data; MEV share may be an underestimate
pbs_snapshot: INFO: relay 'bloxroute_regulated' returned no data; MEV share may be an underestimate
pbs_snapshot: INFO: downloading block data from RPC endpoint
pbs_snapshot: INFO: writing data to 'export/slots_8592501-8593000_snapshot.csv'
pbs_snapshot: INFO: downloading missed slot data from beaconcha.in
All queried relays were available!
Range of blocks analyzed: from 19393170 to 19393629
Number of payloads delivered by relay: ultrasound: 213, flashbots: 120, agnostic: 110, aestus: 24, titan: 1, eden: 1, manifold: 1
Number of payloads delivered by only one relay: ultrasound: 94, flashbots: 86, agnostic: 4, aestus: 3, titan: 1, eden: 1
Share of MEV-Boost blocks: 66.96%
Share of missed slots: 8.00%
Number of proposers that missed their slots: 41
Number of untagged proposers that missed their slots: 5
Number of missed slots by entity: Coinbase: 34, Bitcoin Suisse: 1, Revolut: 1
Analysis of output:
- The requested slots are outside of bloXroute's supported data range, so they are missing from the dataset
- The share of MEV-Boost blocks is underestimated because it is missing blocks only relayed by bloXroute
- The missed slots data is accurate because we calculate the value through the JSON-RPC data
- We see that 34 of 41 missed slots had Coinbase as the expected proposer, implying that they were the source of the outage
(pbs-snapshot) $ ./pbs_snapshot.py -husage: pbs_snapshot.py [-h] [-E] [-l LOOKBACK] [-m MAX_SLOT] [-q] [-r RPC] [-R REQUESTS_PER_SECOND]
Collect data for making assertions about the health of the MEV-Boost relay network
options:
-h, --help show this help message and exit
-E, --no-export-data Do not export CSV data (defaults to false)
-l LOOKBACK, --lookback LOOKBACK
Number of slots by which to look back (defaults to 50)
-m MAX_SLOT, --max-slot MAX_SLOT
Largest slot number to consider for which a block has been added to the chain (defaults to the latest)
-q, --quiet Do not log normal events
-r RPC, --rpc RPC RPC endpoint from which to fetch block metadata (defaults to the value of the ETH_RPC_URL environment variable
or to eth.merkle.io)
-R REQUESTS_PER_SECOND, --requests-per-second REQUESTS_PER_SECOND
Request rate for the given RPC endpoint (defaults to 2)
Some relays may not return any data for some queries. This can happen when the relay was down during the period corresponding to the lookback interval or when the relay no longer has data associated with that interval. In these cases, PBS Snapshot may underestimate the MEV share.
Relay data won't fill the lookback interval when there is at least one missed slot at either endpoint of the lookback interval.
PBS Snapshot is provided under the MIT License.
The included proposer indexes are derived from the Dune Analytics spellbook under the additional use grant.
We acknowledge significant source code and documentation contributions from a frequent but anonymous Data Always collaborator.