add a short connect timeout separate from the overall request timeout.
also fix a bug when only 1 server was in the rpc list causing a very tight loop that made tokio sad
* cargo upgrade --incompatible and update
* first draft at suprisingly_large_refactor_to_get_ids_everywhere
* put app in a task_local
* ref cleanup
* use a OnceLock instead of a tokio local
* test more methods
* APP isn't set in all tests
* it compiles. tests fail. todos still open
* use the app only when necessary
* more tests. less panic
* less verbose debug impl
* short enum names
* move kafka and secrets to their own files
* main tests pass
* add debug chain block time
* helper for stats that ignores internal stats
* Update Jenkinsfile (#223)
* more tests
---------
Co-authored-by: Pewxz <124064710+pewxz@users.noreply.github.com>
* streaming responses compile, still small TODOs & cleanup
* refactor to allow short curcuits on uncacheable requests (#220)
* refactor to allow short curcuits on uncacheable requests
* handle error inside an Arc from moka
* arc instead of box
* lint
* more lint
* add bonus rate limits and keep all semaphores after rate limits (#221)
* add bonus rate limits and keep all semaphores after rate limits
* remove stale todos
* better function names
* cargo upgrade and update
* make some panics warn. more todo
* pass request_metadata through so streaming responses include their final size
* remove stale TODO
* remove more stale todos
* add file i missed
* make to_json_string work well enough
* check config for free_subscriptions
---------
Co-authored-by: Rory Neithinger <rory@llamanodes.com>
* upgrade and start adding deduped_broadcast
* basic test
* accept arg for cache ttl
* first pass at bringing the tx firehose back
* try_send instead of send since there aren't always receivers
* deduped broadcaster needs the try send
* track counts and include in /status
* use config for enable subscribe_txs
* add test_multiple_proxies_stats_add_up
* make a premium user using both proxies
* added a couple clones, must add constraints now to run for multiple-proxies (check arithmetic)
* lint and code review
* fix comment
* fix tests (now fails at todo)
* will introduce endpoint to fetch rpc stats from mysql
* added influxdb to tests, should next do asserst in stats collected by influx, and mysql for multi-proxy
* created test where influx and mysql have separate data, should help with debugging
* forgot to drop influx
* tests pass except multi-proxy
* test passes, will check out nothing broke elswhere
* go back to numbers
* some linting
* linting
* removed redundant info!
* responding to PR comments
* ULID as instance-hash for the tag in influx (for anti-dup)
---------
Co-authored-by: yenicelik <david.yenicelik@gmail.com>
* dont migrate on start. just connect
* it compiles
* that log is loud
* get transaction from a local clone
* import the trait
* put the test app into a thread instead of tokio::spawn
* fix one test
* try db before rpcs
* db connection is too slow. need to wait for it
* do db setup once while spawning
* fix popularity contest
* more info in the Debug for Web3Rpc
* add frontend_requests and cache_misses to the Balance query
* add more to balance and stats flushing and improved test coverage
* it compiles
* deserializer for Ulid to Uuid
I think a wrapper type on Ulid that implements sea_orm::Value is probably better
* rename variable to match struct name
* add deserializer for Address -> Vec<u8>
* sql sum returns a Decimal. need to convert to u64
* assert more
* one log and assert more
* log more
* use a helper to get the user's rpc provider
* this should be 2 now that we have a public and authed call
* this should be zero. the public has the cache miss
* instrument cu calcs
* trace the value we took, not the default that replaced it
* move usd_per_chain into config
* remove some extra logging
* use Arc::into_inner to maybe avoid a race
* off by 1
* pass paid credits used instead of returning it
this lets us use it to write to our user balance cache first. importantly, this keeps us from holding a write lock while writing to mysql
* no cache misses expected in this test
* actually check the admin
* put the balance checks back now that the rest of the test works
* archive request is being set incorrectly
* wow howd we manage flipping the greater than sign on archive depth
* move latest_balance and premium_credits_used to before any stats are emitted
* lint
* and build undoes the linting. fun
i didnt even want to lint them in the first place, so this is fine
* missed incrementing total_spent when not incrementing total_spent_paid_credits
* use the credits on self
* use the credits on self (pt 2)
* fix type for 10 cu query
* convert the requestmetadata on the other side of the channel
* logs
* viewing stats is allowed even without a balance
* move paid_credits_used to AuthorizationChecks
* wip
* test_sum_credits_used finally passes
* UserBalanceCache::get_or_insert
* re-enable rpc_secret_key_cache
* move invalidate to a helper function
and always call it **after** the db is commited
* fix PartialEq and Eq on RpcSecretKey
* cargo upgrade
* WIP will work on locks
* added concurrent test for referral, numbers rn still add up
* will change balance table to be a view instead of a table
* merged in devel. will proceed with migrations for rpc_stats_v2 and referral
* WIP
* WIP
* WIP compiling
* gotta add the balance_v2 view
* about to add balance view
* stupid missing closing param
* compiles again, now i will just have to implement the raw sql
* need to paste the real query and play with it
* ok now on to testing
* addresses many comments in PR
* app works for simple unauthorized access. will look into tests
* some tests pass, referral logic tests fail
* not sure why the test is failing
---------
Co-authored-by: yenicelik <david.yenicelik@gmail.com>
* quick cache and allocate less
* improve /status cache
* prepare to cache raw transaction hashes so we dont dos our backends
* simple benchmark for /health and /status
* mut not needed with atomics
* DRY all the status pages
* use u64 instead of bytes for subscriptions
* fix setting earliest_retry_at and improve logs
* Revert "use kanal instead of flume or tokio channels (#68)"
This reverts commit 510612d343fc51338a8a4282dcc229b50097835b.
* fix automatic retries
* put relaxed back
* convert error message time to seconds
* assert instead of debug_assert while we debug
* ns instead of seconds
* disable peak_latency for now
* null is the default
* cargo fmt
* comments
* remove request caching for now
* log on exit
* unit weigher for now
* make cache smaller. we need a weigher for prod. just debugging
* oops. we need async
* add todo
* no need for to_string on a RawValue
