Prepare v0.77.3 changeset #4648

see the module-level docs of `pd::metrics::sleep_worker`. this introduces a submodule to `pd::metrics`, which observes tokio scheduler latency and records it for the Prometheus exporter. the core idea in this worker is that it will invoke `tokio::time::sleep(..)` to put the task to sleep for one second. upon being woken up by the scheduler, it will check the *new* wall-clock time and calculate the actual amount of time it spent waiting. if this took longer than expected, it increments the `pd_async_sleep_drift_milliseconds` counter by the observed latency, measured in milliseconds. as the module-level docs note, this is a very useful tool to identify when the runtime is being disrupted by blocking I/O or other expensive computation. pd: 🚪 `metrics` module can be `pub` this will make it slightly nicer to expose facilities for telemetry in the `pd` binary, without clouding the top-level namespace. (cherry picked from commit 10421b8)

this commit moves the logic responsible for configuring dex metrics into the dex's metrics module. there are some lurking footguns here, so we can avoid easily forgotten non-local reasoning and move the regex/prefix logic next to the metrics. (cherry picked from commit da1d31e)

Pulls in the `metrics-process` crate [0] to bolt on CPU usage and other OS-level info to the metrics emitted by pd. Doing so will support a better first-run experience for node operators who lack a comprehensive pre-existing setup for monitoring node health, so they can get a sense of how much load pd is under. Adds a cursory first-pass on a new dashboard, but will need to follow up on that once the new metrics actually land on hosts that are receiving load. Also included a reference to metrics addd in #4581, as a treat. [0] https://docs.rs/metrics-process/2.0.0/metrics_process/index.html (cherry picked from commit 3045645)

we can't add counters or gauges named `penumbra_dex_` because of the use of `Matcher::Prefix`. rather, we can tweak this and provide the explicit list of metrics we want buckets for. (cherry picked from commit 2b6621b)

(cherry picked from commit f766aba)

(cherry picked from commit 5c20efe)

## Describe your changes This re-centers the DEX buckets around the expected latency region ## Checklist before requesting a review - [x] If this code contains consensus-breaking changes, I have added the "consensus-breaking" label. Otherwise, I declare my belief that there are not consensus-breaking changes, for the following reason: (cherry picked from commit 06a08f7)

Adds a TimestampByHeightRequest to SCT QueryService. There is no corresponding migration added, so any block heights lower than the height at which this is deployed will have missing data. #4522 - [ ] If this code contains consensus-breaking changes, I have added the "consensus-breaking" label. Otherwise, I declare my belief that there are not consensus-breaking changes, for the following reason: > This adds a pb API endpoint and should not affect consensus. New data is tracked in nonconsensus storage only. --------- Co-authored-by: turbocrime <[email protected]> Co-authored-by: Chris Czub <[email protected]> (cherry picked from commit 989d3a9)

…ailures (#4642) ## Describe your changes The `pcli tx position close-all` and `pcli tx position withdraw-all` commands were failing to plan for me when I had a significant (100+) number of positions. This PR splits them into chunks, currently 30 positions, and fixed the planning failures for me. ## Checklist before requesting a review - [x] If this code contains consensus-breaking changes, I have added the "consensus-breaking" label. Otherwise, I declare my belief that there are not consensus-breaking changes, for the following reason: > pcli only (cherry picked from commit 1693c20)

when a pcli user initializes their configuration, they provide the `init` command with the grpc url of a fullnode, which is stored in pcli's configuration file. by default, this is found at `~/.local/share/pcli/config.toml`. if that fullnode is later encountering issues, this can render many pcli commands unusable, without a clear workaround. this is a small patch, providing a top-level `--grpc-url` option that will override the config file's GRPC url. this can help users temporarily send requests to a different fullnode, until their preferred default comes back online. (cherry picked from commit 3f7ccbd)

This should fix an issue wherein a load balanced RPC can cause data corruption by delivering the same block twice in the stream. ## Issue ticket number and link This should close #4577. ## Checklist before requesting a review - [x] If this code contains consensus-breaking changes, I have added the "consensus-breaking" label. Otherwise, I declare my belief that there are not consensus-breaking changes, for the following reason: > Just a client change (cherry picked from commit 4245ebc)

(cherry picked from commit 516b1a1)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Prepare v0.77.3 changeset #4648

Prepare v0.77.3 changeset #4648

Commits on Jun 20, 2024