* cargo update
* bump lru to 0.19
* bump mockall to 0.15
* bump quanta to 0.13
* bump tikv-jemalloc-ctl and tikv-jemallocator to 0.7
* bump tabled to 0.22
* bump tower-http to 0.7
small breaking change: compression SizeAbove predicate threshold from u16 to u64
* bump ulid to 3.0
Ulid::new() renamed to Ulid::generate()
* bump syn to 3.0 and prettyplease to 0.3
* bump vrl to 0.37
diagnostic feature no longer exists and is now the default
there are some changes in the vrl language wrt to array and NaN handling
* bump multiple opentelemetry deps to 0.33
Bumps opentelemetry, opentelemetry_sdk, opentelemetry-otlp, opentelemetry-appender-tracing to 0.33, tracing-opentelemetry to 0.34 and the pinned metrics-opentelemetry to =0.24033.1.
retries are no longer experimental and are built-in by default
* bump reqwest to 0.13, reqwest-middleware to 0.5, reqwest-retry to 0.9
there are a few breaking changes, feature renamed, function signature
slighly modified
reqwest now transitively depends on platform trust store instead of
webpki-roots, and moved from ring to aws-lc-crypto
* bump opendal and opendal-http-transport-reqwest to 0.59
small breaking change on bulk delete
* dd-rust-license-tool
* restore proper opendal+reqwest+tls test
* use aws-lc instead of ring
we were already doing that in tests, no reason to not unify
* upgrade tokio again
apparently 1.54.0 has an issue with tokio::time::Interval
* always emit generation_id as string
we sometime emited them as string, and sometime as numbers
this made values show up wrong in any system which transformed integers into float
* rename node_id to remote_node_id when referencing a different node
otherwise node_id is ambiguous, sometime the node itself, sometime the node we're talking about
updating memory usage for a permit can drop the permit related usage to zero,
which means the documented invariant is plain wrong, and cause us to reset the
used permit counter when we shouldn't
* estimate current node load
* expose node load as grpc endpoint
* take existing load into account when placing jobs
* ignore node with very high load when computing target load
* handle unimplemented get_load
* improve test
* use atomic instead of actor to maintain/query pending job cost
* don't run a GetLoad call for fetch docs jobs
* complete TODO on compute_split_cost
* add metric for node search load
* add 1s cache to get_load
* rustfmt
* add env to disable load querying
trinity-1686a