When `zeroOrOneCharacterStringMatcher` wach checking the input string,
it assumed that if there are more than one bytes, then there are more
than one runes, but that's not necessarily true.
Signed-off-by: Oleg Zaytsev <mail@olegzaytsev.com>
The strings produced by these tests can run to thousands of characters,
which makes test logs difficult to read.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Restrict the capacity of first argument to `append()` to force an allocation.
This is for the slice implementation only.
Signed-off-by: Domantas Jadenkus <djadenkus@gmail.com>
I was bored on a train and I spent some amount of time trying to scratch
some nanoseconds off the Labels.Compare when running with stringlabels.
I would be ashamed to admit the real amount of time I spent on it.
The worst thing is, I can't really explain why this is performing so
much better, and someone should re-run the benchmarks on their machine
to confirm that it's not something related to general relativity because
the train is moving. I also added some extra real-life benchmark cases
with longer labelsets (these aren't the longest we have in production,
but kubernetes labelsets are fairly common in Prometheus so I thought it
would be nice to have them).
My benchmarks show this diff:
goos: darwin
goarch: arm64
pkg: github.com/prometheus/prometheus/model/labels
│ old │ new │
│ sec/op │ sec/op vs base │
Labels_Compare/equal 5.898n ± 0% 5.875n ± 1% -0.40% (p=0.037 n=10)
Labels_Compare/not_equal 11.78n ± 2% 11.01n ± 1% -6.54% (p=0.000 n=10)
Labels_Compare/different_sizes 4.959n ± 1% 4.906n ± 2% -1.05% (p=0.050 n=10)
Labels_Compare/lots 21.32n ± 0% 17.54n ± 5% -17.75% (p=0.000 n=10)
Labels_Compare/real_long_equal 15.06n ± 1% 14.92n ± 0% -0.93% (p=0.000 n=10)
Labels_Compare/real_long_different_end 25.20n ± 0% 24.43n ± 0% -3.04% (p=0.000 n=10)
geomean 11.86n 11.25n -5.16%
Signed-off-by: Oleg Zaytsev <mail@olegzaytsev.com>
Use a stack buffer to reduce memory allocations.
`Write(AppendQuote(AvailableBuffer` does not allocate or copy when
the buffer has sufficient space.
Also add a benchmark, with some refactoring.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
The individual strings for label names and values are held in a table,
and each Labels value is a run of varint-encoded indexes into that table.
When creating new labels, a sync.Mutex is locked around reads and writes.
When reading labels, there is no locking because the table of strings
used by those labels is immutable.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
The current implementation of `InternStrings` will only save memory
when the whole set of labels is identical to one already seen, and this
cannot happen in the one place it is called from in Prometheus,
remote-write, which already detects identical series.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
This PR is a reference implementation of the proposal described in #10420.
In addition to what described in #10420, in this PR I've introduced labels.StableHash(). The idea is to offer an hashing function which doesn't change over time, and that's used by query sharding in order to get a stable behaviour over time. The implementation of labels.StableHash() is the hashing function used by Prometheus before stringlabels, and what's used by Grafana Mimir for query sharding (because built before stringlabels was a thing).
Follow up work
As mentioned in #10420, if this PR is accepted I'm also open to upload another foundamental piece used by Grafana Mimir query sharding to accelerate the query execution: an optional, configurable and fast in-memory cache for the series hashes.
Signed-off-by: Marco Pracucci <marco@pracucci.com>
This function is called very frequently when executing PromQL functions,
and we can do it much more efficiently inside Labels.
In the common case that `__name__` comes first in the labels, we simply
re-point to start at the next label, which is nearly free.
`DropMetricName` is now so cheap I removed the cache - benchmarks show
everything still goes faster.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
This reduces bulk and should avoid issues if a fix is made in one file
and not the other.
A few methods now call `Range()` instead of `range`, but nothing
performance-sensitive.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
* Labels: reduce allocations when creating from TSDB
When reading the WAL, by passing references into the buffer we can avoid
copying strings under `-tags stringlabels`.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Inline one call to `decodeString`, and skip decoding the value string
until we find a match for the name.
Do a quick check on the first character in each string,
and exit early if we've gone past - labels are sorted in order.
Also improve tests and benchmark:
* labels: test Get with varying lengths - it's not typical for Prometheus labels to all be the same length.
* extend benchmark with label not found
---------
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Instead of unpacking every individual string, we skip to the point
where there is a difference, going 8 bytes at a time where possible.
Add benchmark for Compare; extend tests too.
---------
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Co-authored-by: Oleg Zaytsev <mail@olegzaytsev.com>
This is a minor cosmetical change, but my IDE (and I guess many of them)
nests `labels_string.go` under `labels.go` because it assumes it's the
file generated by the `stringer` tool, which follows that naming
pattern.
Signed-off-by: Oleg Zaytsev <mail@olegzaytsev.com>
labels: dont compile regex matcher if we know its a literal
Signed-off-by: Michael Hoffmann <mhoffm@posteo.de>
Co-authored-by: Sharad <sharadgaur@gmail.com>
* labels: respect Set after Del in Builder
The implementations are not symmetric between `Set()` and `Del()`, so
we must be careful. Add tests for this, both in labels and in relabel
where the issue was reported.
Also make the slice implementation consistent re `slices.Contains`.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Wiser coders than myself have come to the conclusion that a `switch`
statement is almost always superior to a statement that includes any
`else if`.
The exceptions that I have found in our codebase are just these two:
* The `if else` is followed by an additional statement before the next
condition (separated by a `;`).
* The whole thing is within a `for` loop and `break` statements are
used. In this case, using `switch` would require tagging the `for`
loop, which probably tips the balance.
Why are `switch` statements more readable?
For one, fewer curly braces. But more importantly, the conditions all
have the same alignment, so the whole thing follows the natural flow
of going down a list of conditions. With `else if`, in contrast, all
conditions but the first are "hidden" behind `} else if `, harder to
spot and (for no good reason) presented differently from the first
condition.
I'm sure the aforemention wise coders can list even more reasons.
In any case, I like it so much that I have found myself recommending
it in code reviews. I would like to make it a habit in our code base,
without making it a hard requirement that we would test on the CI. But
for that, there has to be a role model, so this commit eliminates all
`if else` occurrences, unless it is autogenerated code or fits one of
the exceptions above.
Signed-off-by: beorn7 <beorn@grafana.com>
Add a fast path for the common case that a string is less than 127 bytes
long, to skip a shift and the loop.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
This is a method used by some downstream projects; it was created to
optimize the implementation in `labels_string.go` but we should have one
for both implementations so the same code works with either.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Deleted labels are remembered, even if they were not in `base` or were
removed from `add`, so `base+add-del` could go negative.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Go spends some time initializing all the elements of these arrays to
zero, so reduce the size from 1024 to 128. This is still much bigger
than we ever expect for a set of labels.
(If someone does have more than 128 labels it will still work, but via
heap allocation.)
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
It took a `Labels` where the memory could be re-used, but in practice
this hardly ever benefitted. Especially after converting `relabel.Process`
to `relabel.ProcessBuilder`.
Comparing the parameter to `nil` was a bug; `EmptyLabels` is not `nil`
so the slice was reallocated multiple times by `append`.
Lastly `Builder.Labels()` now estimates that the final size will depend
on labels added and deleted.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Although we had a different slice, the underlying memory was the same so
any changes meant we could skip some values.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
