This extracts Querier as an instantiateable and closeable object
rather than just defining extending methods of the storage interface.
This improves composability and allows abstracting query transactions,
which can be useful for transaction-level caches, consistent data views,
and encapsulating teardown.
When using the chunking code in other projects (both Weave Prism and
ChronixDB ingester), you sometimes want to know how well you are
utilizing your chunks when closing/storing them.
These more specific methods have replaced `metricForLabelMatchers`
in cases where its `map[fingerprint]metric` result type was
not necessary or was used as an intermediate step
Avoids duplicated calls to `seriesForRange` from
`QueryRange` and `QueryInstant` methods.
This is a followup to https://github.com/prometheus/prometheus/pull/2011.
This publishes more of the methods and other names of the chunk code and
moves the chunk code to its own package. There's some unavoidable
ugliness: the chunk and chunkDesc metrics are used by both packages, so
I had to move them to the chunk package. That isn't great, but I don't
see how to do it better without a larger redesign of everything. Same
for the evict requests and some other types.
* Add config, HTTP Basic Auth and TLS support to the generic write path.
- Move generic write path configuration to the config file
- Factor out config.TLSConfig -> tlf.Config translation
- Support TLSConfig for generic remote storage
- Rename Run to Start, and make it non-blocking.
- Dedupe code in httputil for TLS config.
- Make remote queue metrics global.
This is based on https://github.com/prometheus/prometheus/pull/1997.
This adds contexts to the relevant Storage methods and already passes
PromQL's new per-query context into the storage's query methods.
The immediate motivation supporting multi-tenancy in Frankenstein, but
this could also be used by Prometheus's normal local storage to support
cancellations and timeouts at some point.
CPUs have to serialise write access to a single cache line
effectively reducing level of possible parallelism. Placing
mutexes on different cache lines avoids this problem.
Most gains will be seen on NUMA servers where CPU interconnect
traffic is especially expensive
Before:
go test . -run none -bench BenchmarkFingerprintLocker
BenchmarkFingerprintLockerParallel-4 2000000 932 ns/op
BenchmarkFingerprintLockerSerial-4 30000000 49.6 ns/op
After:
go test . -run none -bench BenchmarkFingerprintLocker
BenchmarkFingerprintLockerParallel-4 3000000 569 ns/op
BenchmarkFingerprintLockerSerial-4 30000000 51.0 ns/op
My aim is to support the new grpc generic write path in Frankenstein. On the surface this seems easy - however I've hit a number of problems that make me think it might be better to not use grpc just yet.
The explanation of the problems requires a little background. At weave, traffic to frankenstein need to go through a couple of services first, for SSL and to be authenticated. So traffic goes:
internet -> frontend -> authfe -> frankenstein
- The frontend is Nginx, and adds/removes SSL. Its done this way for legacy reasons, so the certs can be managed in one place, although eventually we imagine we'll merge it with authfe. All traffic from frontend is sent to authfe.
- Authfe checks the auth tokens / cookie etc and then picks the service to forward the RPC to.
- Frankenstein accepts the reads and does the right thing with them.
First problem I hit was Nginx won't proxy http2 requests - it can accept them, but all calls downstream are http1 (see https://trac.nginx.org/nginx/ticket/923). This wasn't such a big deal, so it now looks like:
internet --(grpc/http2)--> frontend --(grpc/http1)--> authfe --(grpc/http1)--> frankenstein
Next problem was golang grpc server won't accept http1 requests (see https://groups.google.com/forum/#!topic/grpc-io/JnjCYGPMUms). It is possible to link a grpc server in with a normal go http mux, as long as the mux server is serving over SSL, as the golang http client & server won't do http2 over anything other than an SSL connection. This would require making all our service to service comms SSL. So I had a go a writing a grpc http1 server, and got pretty far. But is was a bit of a mess.
So finally I thought I'd make a separate grpc frontend for this, running in parallel with the frontend/authfe combo on a different port - and first up I'd need a grpc reverse proxy. Ideally we'd have some nice, generic reverse proxy that only knew about a map from service names -> downstream service, and didn't need to decode & re-encode every request as it went through. It seems like this can't be done with golang's grpc library - see https://github.com/mwitkow/grpc-proxy/issues/1.
And then I was surprised to find you can't do grpc from browsers! See http://www.grpc.io/faq/ - not important to us, but I'm starting to question why we decided to use grpc in the first place?
It would seem we could have most of the benefits of grpc with protos over HTTP, and this wouldn't preclude moving to grpc when its a bit more mature? In fact, the grcp FAQ even admits as much:
> Why is gRPC better than any binary blob over HTTP/2?
> This is largely what gRPC is on the wire.
This adds a flag -storage.local.engine which allows turning off local
storage in Prometheus. Instead of adding if-conditions and nil checks to
all parts of Prometheus that deal with Prometheus's local storage
(including the web interface), disabling local storage simply means
replacing the normal local storage with a noop version that throws
samples away and returns empty query results. We also don't add the noop
storage to the fanout appender to decrease internal overhead.
Instead of returning empty results, an alternate behavior could be to
return errors on any query that point out that the local storage is
disabled. Not sure which one is more preferable, so I went with the
empty result option for now.
By splitting the single queue into multiple queues and flushing each individual queue in serially (and all queues in parallel), we can guarantee to preserve the order of timestampsin samples sent to downstream systems.
- fold metric name into labels
- return initialization errors back to main
- add snappy compression
- better context handling
- pre-allocation of labels
- remove generic naming
- other cleanups
This uses a new proto format, with scope for multiple samples per
timeseries in future. This will allow users to pump samples out to
whatever they like without having to change the core Prometheus code.
There's also an example receiver to save users figuring out the
boilerplate themselves.
Turns out its valid to have an overall chunk which is smaller than the
full doubleDeltaHeaderBytes size -- if it has a single sample, it
doesn't fill the whole header. Updated unmarshalling check to respect
this.
This is (hopefully) a fix for #1653
Specifically, this makes it so that if the length for the stored
delta/doubleDelta is somehow corrupted to be too small, the attempt to
unmarshal will return an error.
The current (broken) behavior is to return a malformed chunk, which can
then lead to a panic when there is an attempt to read header values.
The referenced issue proposed creating chunks with a minimum length -- I
instead opted to just error on the attempt to unmarshal, since I'm not
clear on how it could be safe to proceed when the length is
incorrect/unknown.
The issue also talked about possibly "quarantining series", but I don't
know the surrounding code well enough to understand how to make that
happen.
Specifically, the TestSpawnNotMoreThanMaxConcurrentSendsGoroutines was failing on a fresh checkout of master.
The test had a race condition -- it would only pass if one of the
spawned goroutines happened to very quickly pull a set of samples off an
internal queue.
This patch rewrites the test so that it deterministically waits until
all samples have been pulled off that queue. In case of errors, it also
now reports on the difference between what it expected and what it found.
I verified that, if the code under test is deliberately broken, the test
successfully reports on that.
See discussion in
https://groups.google.com/forum/#!topic/prometheus-developers/bkuGbVlvQ9g
The main idea is that the user of a storage shouldn't have to deal with
fingerprints anymore, and should not need to do an individual preload
call for each metric. The storage interface needs to be made more
high-level to not expose these details.
This also makes it easier to reuse the same storage interface for remote
storages later, as fewer roundtrips are required and the fingerprint
concept doesn't work well across the network.
NOTE: this deliberately gets rid of a small optimization in the old
query Analyzer, where we dedupe instants and ranges for the same series.
This should have a minor impact, as most queries do not have multiple
selectors loading the same series (and at the same offset).
tl;dr: This is not a fundamental solution to the indexing problem
(like tindex is) but it at least avoids utilizing the intersection
problem to the greatest possible amount.
In more detail:
Imagine the following query:
nicely:aggregating:rule{job="foo",env="prod"}
While it uses a nicely aggregating recording rule (which might have a
very low cardinality), Prometheus still intersects the low number of
fingerprints for `{__name__="nicely:aggregating:rule"}` with the many
thousands of fingerprints matching `{job="foo"}` and with the millions
of fingerprints matching `{env="prod"}`. This totally innocuous query
is dead slow if the Prometheus server has a lot of time series with
the `{env="prod"}` label. Ironically, if you make the query more
complicated, it becomes blazingly fast:
nicely:aggregating:rule{job=~"foo",env=~"prod"}
Why so? Because Prometheus only intersects with non-Equal matchers if
there are no Equal matchers. That's good in this case because it
retrieves the few fingerprints for
`{__name__="nicely:aggregating:rule"}` and then starts right ahead to
retrieve the metric for those FPs and checking individually if they
match the other matchers.
This change is generalizing the idea of when to stop intersecting FPs
and go into "retrieve metrics and check them individually against
remaining matchers" mode:
- First, sort all matchers by "expected cardinality". Matchers
matching the empty string are always worst (and never used for
intersections). Equal matchers are in general consider best, but by
using some crude heuristics, we declare some better than others
(instance labels or anything that looks like a recording rule).
- Then go through the matchers until we hit a threshold of remaining
FPs in the intersection. This threshold is higher if we are already
in the non-Equal matcher area as intersection is even more expensive
here.
- Once the threshold has been reached (or we have run out of matchers
that do not match the empty string), start with "retrieve metrics
and check them individually against remaining matchers".
A beefy server at SoundCloud was spending 67% of its CPU time in index
lookups (fingerprintsForLabelPairs), serving mostly a dashboard that
is exclusively built with recording rules. With this change, it spends
only 35% in fingerprintsForLabelPairs. The CPU usage dropped from 26
cores to 18 cores. The median latency for query_range dropped from 14s
to 50ms(!). As expected, higher percentile latency didn't improve that
much because the new approach is _occasionally_ running into the worst
case while the old one was _systematically_ doing so. The 99th
percentile latency is now about as high as the median before (14s)
while it was almost twice as high before (26s).
If the chunks of a series in the checkpoint are all older then the
latest chunk on disk, the head chunk is persisted and therefore has to
be declared closed.
It would be great to have a test for this, but that would require more
plumbing, subject of #447.
PromQL only requires a much narrower interface than local.Storage in
order to run queries. Narrower interfaces are easier to replace and
test, too.
We could also change the web interface to use local.Querier, except that
we'll probably use appending functions from there in the future.
On Windows, it is not possible to rename or delete a file that is
currerntly open. This change closes the file in dropAndPersistChunks
before it tries to delete it, or rename the temporary file to it.
With a lot of series accessed in a short timeframe (by a query, a
large scrape, checkpointing, ...), there is actually quite a
significant amount of lock contention if something similar is running
at the same time.
In those cases, the number of locks needs to be increased.
On the same front, as our fingerprints don't have a lot of entropy, I
introduced some additional shuffling. With the current state, anly
changes in the least singificant bits of a FP would matter.
But only on DEBUG level.
Also, count and report the two cases of out-of-order timestamps on the
one hand and same timestamp but different value on the other hand
separately.
Before, we checkpointed after every newly detected fingerprint
collision, which is not a problem as long as collisions are
rare. However, with a sufficient number of metrics or particular
nature of the data set, there might be a lot of collisions, all to be
detected upon the first set of scrapes, and then the checkpointing
after each detection will take a quite long time (it's O(n²),
essentially).
Since we are rebuilding the fingerprint mapping during crash recovery,
the previous, very conservative approach didn't even buy us
anything. We only ever read from the checkpoint file after a clean
shutdown, so the only time we need to write the checkpoint file is
during a clean shutdown.
Prometheus is Apache 2 licensed, and most source files have the
appropriate copyright license header, but some were missing it without
apparent reason. Correct that by adding it.
The chunk encoding was hardcoded there because it mostly doesn't
matter what encoding is chosen in that test. Since type 1 is
battle-hardened enough, I'm switching to type 2 here so that we can
catch unexpected problems as a byproduct. My expectation is that the
chunk encoding doesn't matter anyway, as said, but then "unexpected
problems" contains the word "unexpected".
So far, the last sample in a chunk was saved twice. That's required
for adding more samples as we need to know the last sample added to
add more samples without iterating through the whole chunk. However,
once the last sample was added to the chunk before it's full, there is
no need to save it twice. Thus, the very last sample added to a chunk
can _only_ be saved in the header fields for the last sample. The
chunk has to be identifiable as closed, then. This information has
been added to the flags byte.
This improves fuzz testing in two ways:
(1) More realistic time stamps. So far, the most common case in
practice was very rare in the test: Completely regular increases of
the timestamp.
(2) Verify samples by scanning through the whole relevant section of
the series.
For Gorilla-like chunks, this showed two things:
(1) With more regularly increasing time stamps, BenchmarkFuzz is
essentially as fast as with the traditional chunks:
```
BenchmarkFuzzChunkType0-8 2 972514684 ns/op 83426196 B/op 2500044 allocs/op
BenchmarkFuzzChunkType1-8 2 971478001 ns/op 82874660 B/op 2512364 allocs/op
BenchmarkFuzzChunkType2-8 2 999339453 ns/op 76670636 B/op 2366116 allocs/op
```
(2) There was a bug related to when and how the chunk footer is
overwritten to make use for the last sample. This wasn't exposed by
random access as the last sample of a chunk is retrieved from the
values in the header in that case.
This is not a verbatim implementation of the Gorilla encoding. First
of all, it could not, even if we wanted, because Prometheus has a
different chunking model (constant size, not constant time). Second,
this adds a number of changes that improve the encoding in general or
at least for the specific use case of Prometheus (and are partially
only possible in the context of Prometheus). See comments in the code
for details.
It is now also used in label matching, so the name of the metric
changed from `prometheus_local_storage_invalid_preload_requests_total`
to `non_existent_series_matches_total'.
Only return an error where callers are doing something with it except
simply logging and ignoring.
All the errors touched in this commit flag the storage as dirty
anyway, and that fact is logged anyway. So most of what is being
removed here is just log spam.
As discussed earlier, the class of errors that flags the storage as
dirty signals fundamental corruption, no even bubbling up a one-time
warning to the user (e.g. about incomplete results) isn't helping much
because _anything_ happening in the storage has to be doubted from
that point on (and in fact retroactively into the past, too). Flagging
the storage dirty, and alerting on it (plus marking the state in the
web UI) is the only way I can see right now.
As a byproduct, I cleaned up the setDirty method a bit and improved
the logged errors.
WIP: This needs more tests.
It now gets a from and through value, which it may opportunistically
use to optimize the retrieval. With possible future range indices,
this could be used in a very efficient way. This change merely applies
some easy checks, which should nevertheless solve the use case of
heavy rule evaluations on servers with a lot of series churn.
Idea is the following:
- Only archive series that are at least as old as the headChunkTimeout
(which was already extremely unlikely to happen).
- Then maintain a high watermark for the last archival, i.e. no
archived series has a sample more recent than that watermark.
- Any query that doesn't reach to a time before that watermark doesn't
have to touch the archive index at all. (A production server at
Soundcloud with the aforementioned series churn and heavy rule
evaluations spends 50% of its CPU time in archive index
lookups. Since rule evaluations usually only touch very recent
values, most of those lookup should disappear with this change.)
- Federation with a very broad label matcher will profit from this,
too.
As a byproduct, the un-needed MetricForFingerprint method was removed
from the Storage interface.
This finally extracts all the common code of the two chunk iterators
into one. Any future chunk encodings with fast access by index can use
the same iterator by simply providing an indexAccessor. Other future
chunk encodings without fast index access (like Gorilla-style) can
still implement the chunkIterator interface as usual.
For one, remove unneeded methods.
Then, instead of using a channel for all values, use a
bufio.Scanner-like interface. This removes the need for creating a
goroutine and avoids the (unnecessary) locking performed by channel
sending and receiving.
This will make it much easier to write new chunk implementations (like
Gorilla-style encoding).
I needed this today for debugging. It can certainly be improved, but
it's already quite helpful.
I refactored the reading of heads.db files out of persistence, which
is an improvement, too.
I made minor changes to the cli package to allow outputting via the
io.Writer interface.
Obviously, it's really bad to depend on timing here. The proper fix
would be to have something like WaitForIndexing for other things to
wait for, too.
For now, let's see if the wait time increase fixes the issue.
This fixes https://github.com/prometheus/prometheus/issues/1059 , but
not in the obvious way (simply not updating the persist watermark,
because that's actually not that simple - we don't really know what
has gone wrong exactly). As any errors relevant here are most likely
caused by severe and unrecoverable problems with the series file,
Using the now quarantine feature is the right step. We don't really
have to be worried about any inconsistent state of the series because
it will be removed for good ASAP. Another plus is that we don't have
to declare the whole storage dirty anymore.
This requires all the panic calls upon unexpected data to be converted
into errors returned. This pollute the function signatures quite
lot. Well, this is Go...
The ideas behind this are the following:
- panic only if it's a programming error. Data corruptions happen, and
they are not programming errors.
- If we detect a data corruption, we "quarantine" the series,
essentially removing it from the database and putting its data into
a separate directory for forensics.
- Failure during writing to a series file is not considered corruption
automatically. It will call setDirty, though, so that a
crashrecovery upon the next restart will commence and check for
that.
- Series quarantining and setDirty calls are logged and counted in
metrics, but are hidden from the user of the interfaces in
interface.go, whith the notable exception of Append(). The reasoning
is that we treat corruption by removing the corrupted series, i.e. a
query for it will return no results on its next call anyway, so
return no results right now. In the case of Append(), we want to
tell the user that no data has been appended, though.
Minor side effects:
- Now consistently using filepath.* instead of path.*.
- Introduced structured logging where I touched it. This makes things
less consistent, but a complete change to structured logging would
be out of scope for this PR.
Fixes https://github.com/prometheus/prometheus/issues/1401
This remove the last (and in fact bogus) use of BoundaryValues.
Thus, a whole lot of unused (and arguably sub-optimal / ugly) code can
be removed here, too.
In a way, our instants were also ranges, just with the staleness delta
as range length. They are no treated equally, just that in one case,
the range length is set as range, in the other the staleness
delta. However, there are "real" instants where start and and time of
a query is the same. In those cases, we only want to return a single
value (the one closest before or at the equal start and end time). If
that value is the last sample in the series, odds are we have it
already in the series object. In that case, there is no need to pin or
load any chunks. A special singleSampleSeriesIterator is created for
that. This should greatly speed up instant queries as they happen
frequently for rule evaluations.
This implies a slight change of behavior as only samples added to the
respective instance of a memorySeries are returned. However, this is
most likely anyway what we want.
Following cases:
- Server has been restarted: Given the time it takes to cleanly
shutdown and start up a server, the series are now stale anyway. An
improved staleness handling (still to be implemented) will be based
on tracking if a given target is continuing to expose samples for a
given time series. In that case, we need a full scrape cycle to
decide about staleness. So again, it makes sense to consider
everything stale directly after a server restart.
- Series unarchived due to a read request: The series is definitely
stale so we don't want to return anything anyway.
- Freshly created time series or series unarchived because of a sample
append: That happens because appending a sample is imminent. Before
the fingerprint lock is released, the series will have received a
sample, and lastSamplePair will always returned the expected value.
Formalize ZeroSamplePair as return value for non-existing samples.
Change LastSamplePairForFingerprint to return a SamplePair (and not a
pointer to it), which saves allocations in a potentially extremely
frequent call.
This will fix issue #1035 and will also help to make issue #1264 less
bad.
The fundamental problem in the current code:
In the preload phase, we quite accurately determine which chunks will
be used for the query being executed. However, in the subsequent step
of creating series iterators, the created iterators are referencing
_all_ in-memory chunks in their series, even the un-pinned ones. In
iterator creation, we copy a pointer to each in-memory chunk of a
series into the iterator. While this creates a certain amount of
allocation churn, the worst thing about it is that copying the chunk
pointer out of the chunkDesc requires a mutex acquisition. (Remember
that the iterator will also reference un-pinned chunks, so we need to
acquire the mutex to protect against concurrent eviction.) The worst
case happens if a series doesn't even contain any relevant samples for
the query time range. We notice that during preloading but then we
will still create a series iterator for it. But even for series that
do contain relevant samples, the overhead is quite bad for instant
queries that retrieve a single sample from each series, but still go
through all the effort of series iterator creation. All of that is
particularly bad if a series has many in-memory chunks.
This commit addresses the problem from two sides:
First, it merges preloading and iterator creation into one step,
i.e. the preload call returns an iterator for exactly the preloaded
chunks.
Second, the required mutex acquisition in chunkDesc has been greatly
reduced. That was enabled by a side effect of the first step, which is
that the iterator is only referencing pinned chunks, so there is no
risk of concurrent eviction anymore, and chunks can be accessed
without mutex acquisition.
To simplify the code changes for the above, the long-planned change of
ValueAtTime to ValueAtOrBefore time was performed at the same
time. (It should have been done first, but it kind of accidentally
happened while I was in the middle of writing the series iterator
changes. Sorry for that.) So far, we actively filtered the up to two
values that were returned by ValueAtTime, i.e. we invested work to
retrieve up to two values, and then we invested more work to throw one
of them away.
The SeriesIterator.BoundaryValues method can be removed once #1401 is
fixed. But I really didn't want to load even more changes into this
PR.
Benchmarks:
The BenchmarkFuzz.* benchmarks run 83% faster (i.e. about six times
faster) and allocate 95% fewer bytes. The reason for that is that the
benchmark reads one sample after another from the time series and
creates a new series iterator for each sample read.
To find out how much these improvements matter in practice, I have
mirrored a beefy Prometheus server at SoundCloud that suffers from
both issues #1035 and #1264. To reach steady state that would be
comparable, the server needs to run for 15d. So far, it has run for
1d. The test server currently has only half as many memory time series
and 60% of the memory chunks the main server has. The 90th percentile
rule evaluation cycle time is ~11s on the main server and only ~3s on
the test server. However, these numbers might get much closer over
time.
In addition to performance improvements, this commit removes about 150
LOC.
The First time is kind of trivial as we always know it when we create
a new chunkDesc.
The last time is only know when the chunk is closed, so we have to set
it at that time.
The change saves a lot of digging down into the chunk
itself. Especially the last time is relative expensive as it involves
the creation of an iterator. The first time access now doesn't require
locking, which is also a nice gain.
This gives up on the idea to communicate throuh the Append() call (by
either not returning as it is now or returning an error as
suggested/explored elsewhere). Here I have added a Throttled() call,
which has the advantage that it can be called before a whole _batch_
of Append()'s. Scrapes will happen completely or not at all. Same for
rule group evaluations. That's a highly desired behavior (as discussed
elsewhere). The code is even simpler now as the whole ingestion buffer
could be removed.
Logging of throttled mode has been streamlined and will create at most
one message per minute.
Since we are not overestimating the number of chunks to persist
anymore, this commit also adjusts the default value for
-storage.local.memory-chunks. Update of documentation will follow.
