network: faster node shutdown #3416
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ghost
reviewed
Jan 14, 2022
jannotti
added a commit
that referenced
this pull request
Jan 18, 2022
* Three new globals for to help contract-to-contract usability * detritis * Check error * doc comments * Impose limits on the entire "tree" of inner calls. This also increases the realism of testing of multiple app calls in a group by creating the EvalParams with the real constructor, thus getting the pooling stuff tested here without playing games manipulating the ep after construction. * Move appID tracking into EvalContext, out of LedgerForLogic This change increases the seperation between AVM execution and the ledger being used to lookup resources. Previously, the ledger kept track of the appID being executed, to offer a narrower interface to those resources. But now, with app-to-app calls, the appID being executed must change, and the AVM needs to maintain the current appID. * Stupid linter * Fix unit tests error messages * Allow access to resources created in the same transaction group The method will be reworked, but the tests are correct and want to get them visible to team. * Access to apps created in group Also adds some tests that are currently skipped for testing - access to addresses of newly created apps - use of gaid in inner transactions Both require some work to implement the thing being tested. * Remove tracked created mechanism in favor of examining applydata. * Allow v6 AVM code to use in-group created asas, apps (& their accts) One exception - apps can not mutate (put or del) keys from the app accounts, because EvalDelta cannot encode such changes. * lint docs * typo * The review dog needs obedience training. * Use one EvalParams for logic evals, another for apps in dry run We used to use one ep per transaction, shared between sig and and app. But the new model of ep usage is to keep using one while evaluating an entire group. The app ep is now built logic.NewAppEvalParams which, hopefully, will prevent some bugs when we change something in the EvalParams and don't reflect it in what was a "raw" EvalParams construction in debugger and dry run. * Use logic.NewAppEvalParams to decrease copying and bugs in debugger * Simplify use of NewEvalParams. No more nil return when no apps. This way, NewEvalParams can be used for all creations of EvalParams, whether they are intended for logicsig or app use, greatly simplifying the way we make them for use by dry run or debugger (where they serve double duty). * Remove explicit PastSideEffects handling in tealdbg * Always create EvalParams to evaluate a transaction group. We used to have an optimization to avoid creating EvalParams unless there was an app call in the transaction group. But the interface to allow transaction processing to communicate changes into the EvalParams is complicated by that (we must only do it if there is one!) This also allows us to use the same construction function for eps created for app and logic evaluation, simplifying dry-run and debugger. The optimization is less needed now anyway: 1) The ep is now shared for the whole group, so it's only one. 2) The ep is smaller now, as we only store nil pointers instead of larger scratch space objects for non-app calls. * Correct mistaken commit * Spec improvments * More spec improvments, including resource "availability" * Recursively return inner transaction tree * Lint * No need for ConfirmedRound, so don't deref a nil pointer! * license check * Shut up, dawg. * testing: Fix unit test TestAsyncTelemetryHook_QueueDepth (#2685) Fix the unit test TestAsyncTelemetryHook_QueueDepth * Deprecate `FastPartitionRecovery` from `ConsensusParams` (#3386) ## Summary This PR removes `FastPartitionRecovery` option from consensus parameters. The code now acts as if this value is set to true. Closes algorand/go-algorand-internal#1830. ## Test Plan None. * base64 merge cleanup * Remaking a PR for CI (#3398) * Allow setting manager, reserve, freeze, and clawback at goal asset create * Add e2e tests * Add more tests for goal asset create flags Co-authored-by: Fionna <[email protected]> * Remove the extraneous field type arrays. * bsqrt * acct_holding_get, a unified opcode for account field access * Thanks, dawg * [Other] CircleCI pipeline change for binary uploads (#3381) For nightly builds ("rel/nightly"), we want to have deadlock enabled. For rel/beta and rel/stable, we want to make sure we can build and upload a binary with deadlock disabled so that it can be used for release testing and validation purposes. * signer.KeyDilution need not depend on config package (#3265) crypto package need not depend on config. There is an unnecessary dependency on config. signer.KeyDilution takes the `config.ConsensusParams` as argument to pick the DefaultKeyDilution from it. This introduces dependency from the crypto package to config package. Instead, only the DefaultKeyDilution value can be passed to signer.KeyDilution. * CR and more spec simplification * algodump is a tcpdump-like tool for algod's network protocol (#3166) This PR introduces algodump, a tcpdump-like tool for monitoring algod network messages. * Removing C/crypto dependencies from `data/abi` package (#3375) * Feature Networks pipeline related changes (#3393) Added support for not having certain files in signing script * e2e test for inner transaction appls * testing: Add slightly more coverage to TestAcctUpdatesLookupRetry (#3384) Add slightly more coverage to TestAcctUpdatesLookupRetry * add context to (most) agreement logged writes (#3411) Current agreement code only writes a `context : agreement` to a subset of the logged messages. This change extends the said entry, which would make it easier to pre-process logs entries by their corresponding component. The change in this PR is focused on: 1. make sure that the "root" agreement logger always injects the `context : agreement` argument. 2. change the various locations in the agreement code to use the root agreement logger instead of referring to the application-global instance (`logging.Base()`). * network: faster node shutdown (#3416) During the node shutdown, all the current outgoing connections are being disconnected. Since these connections are web sockets, they require a close connection message to be sent. However, sending this message can take a while, and in situations where the other party has already shut down, we might never get a response. That, in turn, would lead the node waiting until the deadline is reached. The current deadline was 5 seconds. This PR changes the deadline during shutdown to be 50ms. * Give max group size * 16 inner txns, regardless of apps present * Adjust test for allowing 256 inners Co-authored-by: Tsachi Herman <[email protected]> Co-authored-by: Tolik Zinovyev <[email protected]> Co-authored-by: Jack <[email protected]> Co-authored-by: Fionna <[email protected]> Co-authored-by: algobarb <[email protected]> Co-authored-by: Shant Karakashian <[email protected]> Co-authored-by: Nickolai Zeldovich <[email protected]> Co-authored-by: Hang Su <[email protected]> Co-authored-by: chris erway <[email protected]>
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
During the node shutdown, all the current outgoing connections are being disconnected.
Since these connections are web sockets, they require a close connection message to be sent.
However, sending this message can take a while, and in situations where the other party has already shut down, we might never get a response. That, in turn, would lead the node waiting until the deadline is reached.
The current deadline was 5 seconds. This PR changes the deadline during shutdown to be 50ms.
Test Plan
Use existing unit tests.