Problem description:
====================
If a rule evaluation referencing a metric/timeseries M happens at a time
when M doesn't have a memory timeseries yet, looking up the fingerprint
for M (via TieredStorage.GetMetricForFingerprint()) will create a new
Metric object for M which gets both: a) attached to a new empty memory
timeseries (so we don't have to ask disk for the Metric's fingerprint
next time), and b) returned to the rule evaluation layer. However, the
rule evaluation layer replaces the name label (and possibly other
labels) of the metric with the name of the recorded rule. Since both
the rule evaluator and the memory storage share a reference to the same
Metric object, the original memory timeseries will now also be
incorrectly renamed.
Fix:
====
Instead of storing a reference to a shared metric object, take a copy of
the object when creating an empty memory timeseries for caching
purposes.
Change-Id: I9f2172696c16c10b377e6708553a46ef29390f1e
So far we've been using Go's native time.Time for anything related to sample
timestamps. Since the range of time.Time is much bigger than what we need, this
has created two problems:
- there could be time.Time values which were out of the range/precision of the
time type that we persist to disk, therefore causing incorrectly ordered keys.
One bug caused by this was:
https://github.com/prometheus/prometheus/issues/367
It would be good to use a timestamp type that's more closely aligned with
what the underlying storage supports.
- sizeof(time.Time) is 192, while Prometheus should be ok with a single 64-bit
Unix timestamp (possibly even a 32-bit one). Since we store samples in large
numbers, this seriously affects memory usage. Furthermore, copying/working
with the data will be faster if it's smaller.
*MEMORY USAGE RESULTS*
Initial memory usage comparisons for a running Prometheus with 1 timeseries and
100,000 samples show roughly a 13% decrease in total (VIRT) memory usage. In my
tests, this advantage for some reason decreased a bit the more samples the
timeseries had (to 5-7% for millions of samples). This I can't fully explain,
but perhaps garbage collection issues were involved.
*WHEN TO USE THE NEW TIMESTAMP TYPE*
The new clientmodel.Timestamp type should be used whenever time
calculations are either directly or indirectly related to sample
timestamps.
For example:
- the timestamp of a sample itself
- all kinds of watermarks
- anything that may become or is compared to a sample timestamp (like the timestamp
passed into Target.Scrape()).
When to still use time.Time:
- for measuring durations/times not related to sample timestamps, like duration
telemetry exporting, timers that indicate how frequently to execute some
action, etc.
*NOTE ON OPERATOR OPTIMIZATION TESTS*
We don't use operator optimization code anymore, but it still lives in
the code as dead code. It still has tests, but I couldn't get all of them to
pass with the new timestamp format. I commented out the failing cases for now,
but we should probably remove the dead code soon. I just didn't want to do that
in the same change as this.
Change-Id: I821787414b0debe85c9fffaeb57abd453727af0f
While a hack, this change should allow us to serve queries
expeditiously during a flush operation.
Change-Id: I9a483fd1dd2b0638ab24ace960df08773c4a5079
Move the stream to an interface, for a number of additional changes
around it are underway.
Conflicts:
storage/metric/memory.go
Change-Id: I4a5fc176f4a5274a64ebdb1cad52600954c463c3
When samples get flushed to disk, they lose sub-second precision anyways. By
already dropping sub-second precision, data fetched from memory vs. disk will
behave the same. Later, we should consider also storing a more compact
representation than time.Time in memory if we're not going to use its full
precision.
The current behavior only adds those samples to the view that are extracted by
the last pass of the last processed op and throws other ones away. This is a
bug. We need to append all samples that are extracted by each op pass.
This also makes view.appendSamples() take an array of samples.
The one-off keys have been replaced with ``model.LabelPair``, which is
indexable. The performance impact is negligible, but it represents
a cognitive simplification.
Some users of GetMetricForFingerprint() end up modifying the returned metric
labelset. Since the memory storage's implementation of
GetMetricForFingerprint() returned a pointer to the metric (and maps are
reference types anyways), the external mutation propagated back into the memory
storage.
The fix is to make a copy of the metric before returning it.
This makes the memory persistence the backing store for views and
adjusts the MetricPersistence interface accordingly. It also removes
unused Get* method implementations from the LevelDB persistence so they
don't need to be adapted to the new interface. In the future, we should
rethink these interfaces.
All staleness and interpolation handling is now removed from the storage
layer and will be handled only by the query layer in the future.
It is the case with the benchmark tool that we thought that we
generated multiple series and saved them to the disk as such, when
in reality, we overwrote the fields of the outgoing metrics via
Go map reference behavior. This was accidental. In the course of
diagnosing this, a few errors were found:
1. ``newSeriesFrontier`` should check to see if the candidate fingerprint is within the given domain of the ``diskFrontier``. If not, as the contract in the docstring stipulates, a ``nil`` ``seriesFrontier`` should be emitted.
2. In the interests of aiding debugging, the raw LevelDB ``levigoIterator`` type now includes a helpful forensics ``String()`` method.
This work produced additional cleanups:
1. ``Close() error`` with the storage stack is technically incorrect, since nowhere in the bowels of it does an error actually occur. The interface has been simplified to remove this for now.
