router: Avoid large up-front allocations#421
Merged
Conversation
We initially opted to pre-allocate the delay queue to avoid allocation at runtime. However, practically, router capacities are _much_ higher than their typical usage; so we incur an extra ~1MB+ of allocation that will never really be used (for outbound routers). This change initializes the router's DelayQueue with only a single slot by default. The capacity will be increased at runtime as needed.
kleimkuhler
reviewed
Jan 31, 2020
Contributor
kleimkuhler
left a comment
There was a problem hiding this comment.
So Cache.values is created with a size of 0 and therefore does not allocate until an item is inserted. Should expirations also just use DelayQueue::default() so that both allocate for the first time when an item is inserted?
Member
Author
|
@kleimkuhler the thought was that we'll ~never create routers that have no services. Or that will be pretty rare. But it's not a particularly important optimization. |
Contributor
Yep makes sense. It was more coming from the fact that we already have an allocation to make when a service is added since |
kleimkuhler
approved these changes
Jan 31, 2020
Contributor
kleimkuhler
left a comment
There was a problem hiding this comment.
Either way though looks good!
hawkw
approved these changes
Jan 31, 2020
olix0r
added a commit
to linkerd/linkerd2
that referenced
this pull request
Feb 4, 2020
This release fixes a bug in the proxy's logging subsystem that could cause the proxy to consume memory until the process is OOMKilled, especially when the proxy was configured to log diagnostic information. The proxy also now properly emits `grpc-status` headers when signaling proxy errors to gRPC clients. This release upgrades the proxy's Rust version, the `http` crate dependency to address RUSTSEC-2019-0033 and RUSTSEC-2019-0034, and the `prost` crate dependency has been patched to address RUSTSEC-2020-02. --- * internal: Introduce a locking middleware (linkerd/linkerd2-proxy#408) * Update to Rust 1.40 with new Cargo.lock format (linkerd/linkerd2-proxy#410) * Update http to v0.1.21 (linkerd/linkerd2-proxy#412) * internal: Split retry, http-classify, and http-metrics (linkerd/linkerd2-proxy#409) * Actually update http to v0.1.21 (linkerd/linkerd2-proxy#413) * patch `prost` 0.5 to pick up security fix (linkerd/linkerd2-proxy#414) * metrics: Make Counter & Gauge atomic (linkerd/linkerd2-proxy#415) * Set grpc-status headers on dispatch errors (linkerd/linkerd2-proxy#416) * trace: update `tracing-subscriber` to 0.2.0-alpha.4 (linkerd/linkerd2-proxy#418) * discover: Warn on discovery error (linkerd/linkerd2-proxy#422) * router: Avoid large up-front allocations (linkerd/linkerd2-proxy#421) * errors: Set correct HTTP version on responses (linkerd/linkerd2-proxy#424) * app: initialize tracing prior to parsing env vars (linkerd/linkerd2-proxy#425) * trace: update tracing-subscriber to 0.2.0-alpha.6 (linkerd/linkerd2-proxy#423)
adleong
pushed a commit
to linkerd/linkerd2
that referenced
this pull request
Feb 4, 2020
This release fixes a bug in the proxy's logging subsystem that could cause the proxy to consume memory until the process is OOMKilled, especially when the proxy was configured to log diagnostic information. The proxy also now properly emits `grpc-status` headers when signaling proxy errors to gRPC clients. This release upgrades the proxy's Rust version, the `http` crate dependency to address RUSTSEC-2019-0033 and RUSTSEC-2019-0034, and the `prost` crate dependency has been patched to address RUSTSEC-2020-02. --- * internal: Introduce a locking middleware (linkerd/linkerd2-proxy#408) * Update to Rust 1.40 with new Cargo.lock format (linkerd/linkerd2-proxy#410) * Update http to v0.1.21 (linkerd/linkerd2-proxy#412) * internal: Split retry, http-classify, and http-metrics (linkerd/linkerd2-proxy#409) * Actually update http to v0.1.21 (linkerd/linkerd2-proxy#413) * patch `prost` 0.5 to pick up security fix (linkerd/linkerd2-proxy#414) * metrics: Make Counter & Gauge atomic (linkerd/linkerd2-proxy#415) * Set grpc-status headers on dispatch errors (linkerd/linkerd2-proxy#416) * trace: update `tracing-subscriber` to 0.2.0-alpha.4 (linkerd/linkerd2-proxy#418) * discover: Warn on discovery error (linkerd/linkerd2-proxy#422) * router: Avoid large up-front allocations (linkerd/linkerd2-proxy#421) * errors: Set correct HTTP version on responses (linkerd/linkerd2-proxy#424) * app: initialize tracing prior to parsing env vars (linkerd/linkerd2-proxy#425) * trace: update tracing-subscriber to 0.2.0-alpha.6 (linkerd/linkerd2-proxy#423)
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
We initially opted to pre-allocate the delay queue to avoid allocation
at runtime. However, practically, router capacities are much higher
than their typical usage; so we incur an extra ~1MB+ of allocation that
will never really be used (for outbound routers).
This change initializes the router's DelayQueue with only a single slot
by default. The capacity will be increased at runtime as needed.