Skip to main content

Provider Rewards

πŸ’° Earning Rewards​

After staking on chain, providers serve consumers and get cryptographic proofs of relay service. These proofs are submitted on chain to accumulate CUs that lead to token rewards. The calculation for provider rewards is complex and contingent on the validity of submitted proofs. However, given valid proofs, rewards are mostly affected by 3 factors:

  • β˜‘οΈ Passable Quality of Service - The individual score of a relay session. Passable QoS directly alters rewards in a given session.
  • ⭐ Quality of Service Excellence - The reputational score of a provider. QoS Excellence affects provider selection both within a pairing and in the pairing probabilities.
  • 🚨 Jail - The disabling mechanism for inferior/defunct providers. Jailing prevents a provider from giving service for a period of time.

πŸ“Š Passable QoS​

Passable Quality of Service is scored separately in each relay session. Lower scores mean lower rewards. Up to half the accumulated CU can be reduced for bad service. Passable QoS metrics can be viewed both in the Lava Info explorer and Prometheus metrics.

Passable QoS is binary, either it's good or bad there is no in-between. Scores in the range 0-1 are a result of averaging that binary score across relays. You can learn more about Passable QoS from our πŸ“„ RSCH-1000 research paper.

Metrics πŸ“ˆβ€‹

Passable Quality of Service divides into three metrics: Availability, Sync/Freshness of data, Latency.

πŸ›ŽοΈ Availability​

Score 0 or 1 per relay, averaged across relays in a session to give a range 0-1 for each session. 0 is given for each failed relay. A lower availability score can be the result of failed relays by one of the following:

Low-Score Causes​
❗time out
⚠️ The provider does not respond in the maximum allowed time, and the consumer gives up.
βœ… Timeout can be checked for in the provider logs, requests that time out are also identified by the provider and terminated
❗ unsupported chain
⚠️ The provider runs with disabled chains βœ… It can be checked in the provider initialization logs and on the request logs, also on the prometheus metrics for disabled chains
❗ errors
⚠️ Any lava error, whether unsynced epoch in the provider, wrong wallet, misconfigured url
βœ… Error can be checked in the provider logs, and prometheus metrics for disabled chains

Improving availability involves finding the cause of errors, and taking the necessary actions to resolve them.

⚑ add more endpoints
lava supports setting more than one endpoint, and consumers know to loadbalance in case one is down, by setting several endpoints (with several services) providers can avoid downtime

⏲️ Latency​

Score 0 or 1 per relay, averaged across relays in a session to give a range 0-1 per session. 0 is given for each relay that took above half the timeout. A lower latency score can be the result of slow responses, and can be identified by turning on debug logs in the provider to see the latency or checking Prometheus.

⚑ activate lavap cache
Caching can boost up performance and latency at the expense of some memory, if you didn't set a valid cache configuration it could reduce latency scores. Visit the provider features page to ensure caching is correctly enabled.
⚑ set the provider service in proximity to the node
a provider service that is located in the same server or vpc as the node will provide the best latency, while a provider that has to go through TLS and connect to a different location will have increased latency
⚑ set the correct geolocation
lava matches consumers and providers based on geolocation, if your services do not cover several locations with several nodes, it is recommended to set geolocations that are in proximity to your provider service and node. if you misconfigure geolocations that are far from your node's location, your latency scores will be lower
⚑ reduce server load or improve the machine
servers that receive too much usage to the machine capabilities will respond slower, it is recommended to monitor CPU, networking and memory usage
⚑ add more endpoints
lava supports setting more than one endpoint, and consumers know to pick the closest one, by setting more endpoints latency scores will be lower

🌿 Sync/freshness​

Score 0 or 1 per relay, averaged across relays in a session to give a range 0-1 per session. A 0 is given for each relay that has a latest block proof that is older than the spec allowed block lag for QoS sync.

How to Identify​

The freshness proofs are updated by the provider service in a GET_BLOCKNUM request, and then returned together with consumer relay responses. It is possible to turn on debug logs and see the blocks advancing. The latest block is exported in prometheus, and also uploaded on chain. You can compare the latest block on your provider to other providers, a useful way to do this is filter provider_latest_block_report and compare your results to others:

lavap test events 2000 --event lava_provider_latest_block_report --node
Low-Score Causes​
❗ provider state is not updating fast enough
make sure GET_BLOCKNUM requests do not fail, common problems are throttling, or a url misconfiguration
❗ latest block is too old
some nodes can lose sync and return stale data, it is imperative to monitor these cases and fix them, making sure the node is synced

πŸ“Š QoS Excellence​

QoS Excellence is calculated very similarly to Passable QoS. QoS Excellence provides a range of scores that are time-weighted to take the latest information all the actions mentioned here to improve passable QoS affect excellence

Metrics πŸ“ˆβ€‹

Excellence Quality of score divides into 3 metrics:

  • Availability - score in the range 0-1
  • Sync/ Freshness of data - how much time behind other providers are we, lower is better, 0 means your sync is the best in the pairing
  • Latency - how many benchmark ticks passed during a relay in average (time taken / benchmark time). lower is better

πŸ›οΈ Jailing​

How to Fix Getting Jailed πŸ”“β€‹

Lava Protocol removes providers that are providing inferior service. The mechanism is detached from QoS measurements. In order to avoid being jailed, a provider needs to avoid the following:

  • have the staked endpoint not respond to connections
  • have disabled chains in the staked endpoint
  • have too many consecutive errors with a large group of consumers
  • have a non TLS connection or an expired certificate
  • block headers or origins
  • miss on getting rewards

How Jailing Happens ❓​

Once one or more of the aforementioned conditions are met, Lava's Blockchain jails a provider if:

  • there are enough other providers in the spec
  • the provider is not frozen (if you freeze for maintenance you will not get jailed)
  • the provider is active for at least 8 epochs
  • in the last 8 epochs the provider got less sum of rewards than reports sum in the last 2 epochs

Identify Getting Jailed πŸ©»β€‹

If your provider got jailed it will stop receiving requests upon the next epoch.

It is possible to monitor this event via info webpage or the following commands :

⌨️ command when being reported before getting jailed:

lavap test events 2000 --event lava_provider_reported --node

a group of reports can lead to being jailed.

⌨️ command when the blockchain's criteria for jail are met:

lavap test events 2000 --event lava_provider_jailed --node

Resuming Service ▢️​

Since version 0.27.0, providers that are jailed can resume service by unfreezing. It is planned to have a cooldown period on unfreezing.

If repeated jails are activated in a short period:

lavad tx pairing unfreeze --help

Reasons βš–οΈβ€‹

Jail reports contain additional info on the report reason and they can be either due to:

  • disconnections - a provider did not respond to connection attempts
  • errors - a provider's responses were a sequence of consecutive errors.

In addition, the reports contain an exact time tag, so the provider can check the events:

lavap test events 2000 --event lava_provider_reported --node

Disconnections πŸ’’β€‹

These mean the provider's endpoint did not respond and can be due to the following problems:

  • TLS certificate outdated or not set
  • misconfiguration proxying the requests to the provider service
  • provider service not running
  • wrong endpoint in the stake entry on chain: can be fixed by running
lavad tx pairing modify-provider ${CHAIN} --endpoints "${ENDPOINTS}" --geolocation ${GEOLOCATION} --from ${WALLET}

Make sure the url in the endpoint is the provider grpc listening address and not your node url

⏺️ Identifying a Disconnect​

Disconnection problems can be identified by running the test command:

lavap test rpcprovider ${PUBLIC_ADDRESS}

Errors βŒβ€‹

these mean the provider service connection was solid but all relays turned to be errors, might be caused by the following:

  • disabled chain - the provider doesnt have access to the node, or a verification does not pass, and the chain is disabled
  • unexpected errors
  • timeouts

⏺️ Identifying an Error​

Errors can be identified by looking at the provider service logs, it is recommended to run with debug if repeatedly getting jailed.