This changes the benchmarks to be less heavy weight.
It allows the baseline to have no allocations and be
very lightweight. This makes it easier to see
realistic numbers for the costs of the various
operations, making it easier to track and optimize.
Change-Id: I07699881d6976c3ab3432373f057b563752509b2
Reviewed-on: https://go-review.googlesource.com/c/tools/+/225337
Run-TryBot: Ian Cottrell <iancottrell@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
This separates the concerns of tag collections that have to be iterated
and tag collections that need lookup by key.
Also make it so that events just carry a plain slice of tags.
We pass a TagMap down through the exporters and allow it to be
extended on the way.
We no longer need the event.Query method (or the event type)
We now exclusivley use Key as the identity, and no longer have a
common core implementation but just implement it directly in each
type.
This removes some confusion that was causing the same key through
different paths to end up with a different identity.
Change-Id: I61e47adcb397f4ca83dd90342b021dd8e9571ed3
Reviewed-on: https://go-review.googlesource.com/c/tools/+/224278
Run-TryBot: Ian Cottrell <iancottrell@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
This changes the way keys work, there is still a single internal key
implementation for performance reasons, but the public interface is a set of key
implementations that have type safe Of and Get methods.
This also hides the implemenation of Tag so that we can modify the storage form
and find a more efficient storage if needed.
Change-Id: I6a39cc75c2824c6a92e52d59f16e82e876f16e9c
Reviewed-on: https://go-review.googlesource.com/c/tools/+/223137
Run-TryBot: Ian Cottrell <iancottrell@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
This allows early exporters to adjust the event for later ones.
This is used to lookup key values from the context if needed.
Also add a Query type event which is intended to perform all event
modifications but nothing else, and is used to lookup values from
the context. This cleans up a weirdness where the current lookup
presumes there will be an exporter with a matching mechanism.
Change-Id: I835d1e0b2511553c30f94b7becfe7b7b5462c111
Reviewed-on: https://go-review.googlesource.com/c/tools/+/223657
Run-TryBot: Ian Cottrell <iancottrell@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
This makes keys a pointer, which reduces the allocations during logging, and has
a significant improvement in memory and cpu cost when there is no exporter.
Packing a string into an interface requires a 16 byte allocation, packing a pointer
does not.
name old time/op new time/op delta
/LogNoExporter-8 4.39µs ± 2% 3.83µs ± 2% -12.74% (p=0.000 n=18+20)
/Log-8 23.5µs ± 2% 22.6µs ± 1% -4.20% (p=0.000 n=19+18)
name old alloc/op new alloc/op delta
/LogNoExporter-8 1.70kB ± 0% 1.38kB ± 0% -18.78% (p=0.000 n=20+20)
/Log-8 4.91kB ± 0% 4.91kB ± 0% ~ (p=1.000 n=20+20)
name old allocs/op new allocs/op delta
/LogNoExporter-8 83.0 ± 0% 63.0 ± 0% -24.10% (p=0.000 n=20+20)
/Log-8 163 ± 0% 163 ± 0% ~ (all equal)
Change-Id: Iec127f1bff8d5c8f4bd0a6d9f6d8fc4b8bc740b2
Reviewed-on: https://go-review.googlesource.com/c/tools/+/222599
Run-TryBot: Ian Cottrell <iancottrell@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
This is now the only package that is exposed to normal use, and should
be the only thing to appear in libraries.
Change-Id: I90ee47c6519f30db16ff5d5d2910be86e91e5df2
Reviewed-on: https://go-review.googlesource.com/c/tools/+/222557
Run-TryBot: Ian Cottrell <iancottrell@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
We no longer use the span as the core type of tracing, instead that is an
artifact of the exporter, and start and end tracing is just event based.
This both makes the interface normalized, and also means the null exporter case
is considerably cheaper in memory and cpu.
See below for benchstat changes
name old time/op new time/op delta
TracingNoExporter-8 4.19µs ±12% 2.71µs ±11% -35.33% (p=0.000 n=20+20)
Tracing-8 24.1µs ± 3% 5.1µs ±17% -78.66% (p=0.000 n=16+20)
name old alloc/op new alloc/op delta
TracingNoExporter-8 2.32kB ± 0% 0.40kB ± 0% -82.76% (p=0.000 n=20+20)
Tracing-8 6.32kB ± 0% 2.32kB ± 0% -63.30% (p=0.000 n=20+20)
name old allocs/op new allocs/op delta
TracingNoExporter-8 35.0 ± 0% 15.0 ± 0% -57.14% (p=0.000 n=20+20)
Tracing-8 215 ± 0% 35 ± 0% -83.72% (p=0.000 n=20+20)
Change-Id: I3cf25871fa49584819504b5c19aa580e5dd03395
Reviewed-on: https://go-review.googlesource.com/c/tools/+/221740
Run-TryBot: Ian Cottrell <iancottrell@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
These are just like the logging benchmarks but for tracing instead.
Also makes the log writer write out tracing events as well if it is
not in only errors mode
Change-Id: Ie00d7c80f7e2b9433787603261950f70ab1c1e9d
Reviewed-on: https://go-review.googlesource.com/c/tools/+/221739
Run-TryBot: Ian Cottrell <iancottrell@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
