| Age | Commit message (Collapse) | Author |
|---|
|
* Tier part 1
* Tier part 2
* Tier part 3
* Tier part 4
* Tier part 5
* Fix some ML compilation errors
* fix more conflicts
* pass proper tier
* move metric_uuid from state to RRDDIM
* move aclk_live_status from state to RRDDIM
* move ml_dimension from state to RRDDIM
* abstracted the data collection interface
* support flushing for mem db too
* abstracted the query api
* abstracted latest/oldest time per metric
* cleanup
* store_metric for tier1
* fix for store_metric
* allow multiple tiers, more than 2
* state to tier
* Change storage type in db. Query param to request min, max, sum or average
* Store tier data correctly
* Fix skipping tier page type
* Add tier grouping in the tier
* Fix to handle archived charts (part 1)
* Temp fix for query granularity when requesting tier1 data
* Fix parameters in the correct order and calculate the anomaly based on the anomaly count
* Proper tiering grouping
* Anomaly calculation based on anomaly count
* force type checking on storage handles
* update cmocka tests
* fully dynamic number of storage tiers
* fix static allocation
* configure grouping for all tiers; disable tiers for unittest; disable statsd configuration for private charts mode
* use default page dt using the tiering info
* automatic selection of tier
* fix for automatic selection of tier
* working prototype of dynamic tier selection
* automatic selection of tier done right (I hope)
* ask for the proper tier value, based on the grouping function
* fixes for unittests and load_metric_next()
* fixes for lgtm findings
* minor renames
* add dbengine to page cache size setting
* add dbengine to page cache with malloc
* query engine optimized to loop as little are required based on the view_update_every
* query engine grouping methods now do not assume a constant number of points per group and they allocate memory with OWA
* report db points per tier in jsonwrap
* query planer that switches database tiers on the fly to satisfy the query for the entire timeframe
* dbegnine statistics and documentation (in progress)
* calculate average point duration in db
* handle single point pages the best we can
* handle single point pages even better
* Keep page type in the rrdeng_page_descr
* updated doc
* handle future backwards compatibility - improved statistics
* support &tier=X in queries
* enfore increasing iterations on tiers
* tier 1 is always 1 iteration
* backfilling higher tiers on first data collection
* reversed anomaly bit
* set up to 5 tiers
* natural points should only be offered on tier 0, except a specific tier is selected
* do not allow more than 65535 points of tier0 to be aggregated on any tier
* Work only on actually activated tiers
* fix query interpolation
* fix query interpolation again
* fix lgtm finding
* Activate one tier for now
* backfilling of higher tiers using raw metrics from lower tiers
* fix for crash on start when storage tiers is increased from the default
* more statistics on exit
* fix bug that prevented higher tiers to get any values; added backfilling options
* fixed the statistics log line
* removed limit of 255 iterations per tier; moved the code of freezing rd->tiers[x]->db_metric_handle
* fixed division by zero on zero points_wanted
* removed dead code
* Decide on the descr->type for the type of metric
* dont store metrics on unknown page types
* free db_metric_handle on sql based context queries
* Disable STORAGE_POINT value check in the exporting engine unit tests
* fix for db modes other than dbengine
* fix for aclk archived chart queries destroying db_metric_handles of valid rrddims
* fix left-over freez() instead of OWA freez on median queries
Co-authored-by: Costa Tsaousis <costa@netdata.cloud>
Co-authored-by: Vladimir Kobal <vlad@prokk.net>
|
|
* netdata doubles
* fix cmocka test
* fix cmocka test again
* fix left-overs of long double to NETDATA_DOUBLE
* RRDDIM detached from disk representation; db settings in [db] section of netdata.conf
* update the memory before saving
* rrdset is now detached from file structures too
* on memory mode map, update the memory mapped structures on every iteration
* allow RRD_ID_LENGTH_MAX to be changed
* granularity secs, back to update every
* fix formatting
* more formatting
|
|
make ml configuration example cleare
|
|
* set grouping functions
* storage engine should check the validity of timestamps, not the query engine
* calculate and store in RRDR anomaly rates for every query
* anomaly rate used by volume metric correlations
* mc volume should use absolute data, to avoid cancelling effect
* return anomaly-rates in jasonwrap with jw-anomaly-rates option to data queries
* dont return null on anomaly rates
* allow passing group query options from the URL
* added countif to the query engine and used it in metric correlations
* fix configure
* fix countif and anomaly rate percentages
* added group_options to metric correlations; updated swagger
* added newline at the end of yaml file
* always check the time the highlighted window was above/below the highlighted window
* properly track time in memory queries
* error for internal checks only
* moved pack_storage_number() into the storage engines
* moved unpack_storage_number() inside the storage engines
* remove old comment
* pass unit tests
* properly detect zero or subnormal values in pack_storage_number()
* fill nulls before the value, not after
* make sure math.h is included
* workaround for isfinite()
* fix for isfinite()
* faster isfinite() alternative
* fix for faster isfinite() alternative
* next_metric() now returns end_time too
* variable step implemented in a generic way
* remove left-over variables
* ensure we always complete the wanted number of points
* fixes
* ensure no infinite loop
* mc-volume-improvements: Add information about invalid condition
* points should have a duration in the past
* removed unneeded info() line
* Fix unit tests for exporting engine
* new_point should only be checked when it is fetched from the db; better comment about the premature breaking of the main query loop
Co-authored-by: Thiago Marques <thiagoftsm@gmail.com>
Co-authored-by: Vladimir Kobal <vlad@prokk.net>
|
|
This will enable first set of initial models to be trained quicker and makes sense now that ml is enabled by default.
|
|
* enabled ml by default
|
|
|
|
|
|
Ideally, we'd log such cases but this is not currently feasible because
we have to process thousands of dimensions per second.
|
|
|
|
|
|
* initial version of worker utilization
* working example
* without mutexes
* monitoring DBENGINE, ACLKSYNC, WEB workers
* added charts to monitor worker usage
* fixed charts units
* updated contexts
* updated priorities
* added documentation
* converted threads to stacked chart
* One query per query thread
* Revert "One query per query thread"
This reverts commit 6aeb391f5987c3c6ba2864b559fd7f0cd64b14d3.
* fixed priority for web charts
* read worker cpu utilization from proc
* read workers cpu utilization via /proc/self/task/PID/stat, so that we have cpu utilization even when the jobs are too long to finish within our update_every frequency
* disabled web server cpu utilization monitoring - it is now monitored by worker utilization
* tight integration of worker utilization to web server
* monitoring statsd worker threads
* code cleanup and renaming of variables
* contrained worker and statistics conflict to just one variable
* support for rendering jobs per type
* better priorities and removed the total jobs chart
* added busy time in ms per job type
* added proc.plugin monitoring, switch clock to MONOTONIC_RAW if available, global statistics now cleans up old worker threads
* isolated worker thread families
* added cgroups.plugin workers
* remove unneeded dimensions when then expected worker is just one
* plugins.d and streaming monitoring
* rebased; support worker_is_busy() to be called one after another
* added diskspace plugin monitoring
* added tc.plugin monitoring
* added ML threads monitoring
* dont create dimensions and charts that are not needed
* fix crash when job types are added on the fly
* added timex and idlejitter plugins; collected heartbeat statistics; reworked heartbeat according to the POSIX
* the right name is heartbeat for this chart
* monitor streaming senders
* added streaming senders to global stats
* prevent division by zero
* added clock_init() to external C plugins
* added freebsd and macos plugins
* added freebsd and macos to global statistics
* dont use new as a variable; address compiler warnings on FreeBSD and MacOS
* refactored contexts to be unique; added health threads monitoring
Co-authored-by: Stelios Fragkakis <52996999+stelfrag@users.noreply.github.com>
|
|
|
|
* rrd: move API structures out of rrddim_volatile
In C, unlike C++, it's not possible to reference a nested structure
from outside this structure.
Since we later want to use rrddim_query_ops and rrddim_collect_ops
separately from rrddim_volatile, move these nested structures out.
* rrd: use opaque handle types for different memory modes
|
|
- set `dbengine anomaly rate every` to 30 by default to provide better sorting functionality for anomaly advisor.
- increase upper clamp on `MaxTrainSamples` to 24 hours for those who might like a larger training window on a parent.
- decrease lower clamp on `MinTrainSamples` to 15 minutes to enable faster training of initial models.
|
|
expand on the various parent-child config options
|
|
Originally, the main training/detection thread loops where meant to be
run only for the localhost host. They would stop when `netdata_exit` was set
to true during the shutdown process.
By enabling training/detection for children, we have to explicitly
cancel ML threads because the service thread can free a child host at any
point in time without setting `netdata_exit` to true.
To support this:
- We send a cancellation request to the training and the detection
threads when we call rrdhost_free.
- We disable/enable cancelation for the actual training/detection
step on every iteration (in order to protect locks and shared
data structures).
|
|
Initialization of ML-related structures and threads should happen
when the underlying RRD objects have been fully initialized. Destruction
should happen in the opposite way, ie. before deleting an RRD host/dimension.
|
|
Co-authored-by: Vasilis Kalintiris <vasilis@netdata.cloud>
|
|
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
|
|
|
|
|
|
* Clamp LagN to non-zero values.
* Free static threads even on test failure.
* Initialize rusage.
* s/free/freez/
|
|
* Move CPU usage stats under netdata charts
Use the hostname in each chart's name, and the machine GUID in each
chart's id.
* Move anomaly_detection.* charts to child host instance.
* Add option to enable/disable streaming of ML-related charts.
* Update priority of prediction/training charts.
|
|
* Increase training thread's max sleep time.
With this change we will only cap the allotted time when it is more than
ten seconds. The previous limit was one second, which had the effect of
scheduling dimensions near the beggining of each training window. This
was not desirable because it would cause high CPU usage on parents with
many children.
* Only exclude netdata.* charts from training.
* Use heartbeat in detection thread.
* Track rusage of prediction thread.
* Track rusage of training thread.
* Add support for random sampling of extracted features.
* Rebase
* Skip RNG when ML is disabled and fix undef behaviour
|
|
|
|
* initial setting up of notebook
* add open in colab button
* draft work
* first version of notebook
* fix open in colab button
* Update ml/notebooks/README.md
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
* use underscores in filename
* add one final visualization approach using scatter plots
* get a better random sample for plots
* small text update
* fix link
* Update ml/notebooks/netdata_anomaly_detection_deepdive.ipynb
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
* Update ml/notebooks/netdata_anomaly_detection_deepdive.ipynb
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
* Update ml/notebooks/netdata_anomaly_detection_deepdive.ipynb
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
* Update ml/notebooks/netdata_anomaly_detection_deepdive.ipynb
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
* Update ml/notebooks/netdata_anomaly_detection_deepdive.ipynb
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
* Update ml/notebooks/netdata_anomaly_detection_deepdive.ipynb
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
* address review comments
* add ipynb files to dockerignore
Co-authored-by: Tina Luedtke <kickoke@users.noreply.github.com>
|
|
|
|
FE needs this information when they issue a `/data` request on the
anomaly rates chart. However, this information is only available at
the creation time of the anomaly rate dimension.
|
|
|
|
* Track anomaly rates with DBEngine.
This commit adds support for tracking anomaly rates with DBEngine. We
do so by creating a single chart with id "anomaly_detection.anomaly_rates" for
each trainable/predictable host, which is responsible for tracking the anomaly
rate of each dimension that we train/predict for that host.
The rrdset->state->is_ar_chart boolean flag is set to true only for anomaly
rates charts. We use this flag to:
- Disable exposing the anomaly rates charts through the functionality
in backends/, exporting/ and streaming/.
- Skip generation of configuration options for the name, algorithm,
multiplier, divisor of each dimension in an anomaly rates chart.
- Skip the creation of health variables for anomaly rates dimensions.
- Skip the chart/dim queue of ACLK.
- Post-process the RRDR result of an anomaly rates chart, so that we can
return a sorted, trimmed number of anomalous dimensions.
In a child/parent configuration where both the child and the parent run
ML for the child, we want to be able to stream the rest of the ML-related
charts to the parent. To be able to do this without any chart name collisions,
the charts are now created on localhost and their IDs and titles have the node's
machine_guid and hostname as a suffix, respectively.
* Fix exporting_engine tests.
* Restore default ML configuration.
The reverted changes where meant for local testing only. This commit
restores the default values that we want to have when someone runs
anomaly detection on their node.
* Set context for anomaly_detection.* charts.
* Check for anomaly rates chart only with a valid pointer.
* Remove duplicate code.
* Use a more descriptive name for id/title pair variable
|
|
Detect dimensions whose values do not change, and skip them from
training. This allows us to reduce the number of training operations
by ~40-50%.
Notice that we don't skip the very 1st training iteration, because a
dimension's value might change at any point in time, and we need to
have a trained model in order to compute its anomaly score.
|
|
Use absolute features when doing training/prediction.
|
|
|
|
* Provide runtime ml info from a new endpoint.
* Add hosts & charts to skip from ML in the /info endpoint.
This information belongs in /info, and not in /ml-info, because the
value of these variables can not change at the agent's runtime.
* Use strdupz instead of strdup.
|
|
Create README.md for `netdata/ml`
|
|
|
|
|
|
* Add support for feature extraction and K-Means clustering.
This patch adds support for performing feature extraction and running the
K-Means clustering algorithm on the extracted features.
We use the open-source dlib library to compute the K-Means clustering
centers, which has been added as a new git submodule.
The build system has been updated to recognize two new options:
1) --enable-ml: build an agent with ml functionality, and
2) --enable-ml-tests: support running tests with the `-W mltest`
option in netdata.
The second flag is meant only for internal use. To build tests successfully,
you need to install the GoogleTest framework on your machine.
* Boilerplate code to track hosts/dims and init ML config options.
A new opaque pointer field is added to the database's host and dimension
data structures. The fields point to C++ wrapper classes that will be used
to store ML-related information in follow-up patches.
The ML functionality needs to iterate all tracked dimensions twice per
second. To avoid locking the entire DB multiple times, we use a
separate dictionary to add/remove dimensions as they are created/deleted
by the database.
A global configuration object is initialized during the startup of the
agent. It will allow our users to specify ML-related configuration
options, eg. hosts/charts to skip from training, etc.
* Add support for training and prediction of dimensions.
Every new host spawns a training thread which is used to train the model
of each dimension.
Training of dimensions is done in a non-batching mode in order to avoid
impacting the generated ML model by the CPU, RAM and disk utilization of
the training code itself.
For performance reasons, prediction is done at the time a new value
is pushed in the database. The alternative option, ie. maintaining a
separate thread for prediction, would be ~3-4x times slower and would
increase locking contention considerably.
For similar reasons, we use a custom function to unpack storage_numbers
into doubles, instead of long doubles.
* Add data structures required by the anomaly detector.
This patch adds two data structures that will be used by the anomaly
detector in follow-up patches.
The first data structure is a circular bit buffer which is being used to
count the number of set bits over time.
The second data structure represents an expandable, rolling window that
tracks set/unset bits. It is explicitly modeled as a finite-state
machine in order to make the anomaly detector's behaviour easier to test
and reason about.
* Add anomaly detection thread.
This patch creates a new anomaly detection thread per host. Each thread
maintains a BitRateWindow which is updated every second based on the
anomaly status of the correspondent host.
Based on the updated status of the anomaly window, we can identify the
existence/absence of an anomaly event, it's start/end time and the
dimensions that participate in it.
* Create/insert/query anomaly events from Sqlite DB.
* Create anomaly event endpoints.
This patch adds two endpoints to expose information about anomaly
events. The first endpoint returns the list of anomalous events within a
specified time range. The second endpoint provides detailed information
about a single anomaly event, ie. the list of anomalous dimensions in
that event along with their anomaly rate.
The `anomaly-bit` option has been added to the `/data` endpoint in order
to allow users to get the anomaly status of individual dimensions per
second.
* Fix build failures on Ubuntu 16.04 & CentOS 7.
These distros do not have toolchains with C++11 enabled by default.
Replacing nullptr with NULL should be fix the build problems on these
platforms when the ML feature is not enabled.
* Fix `make dist` to include ML makefiles and dlib sources.
Currently, we add ml/kmeans/dlib to EXTRA_DIST. We might want to
generate an explicit list of source files in the future, in order to
bring down the generated archive's file size.
* Small changes to make the LGTM & Codacy bots happy.
- Cast unused result of function calls to void.
- Pass a const-ref string to Database's constructor.
- Reduce the scope of a local variable in the anomaly detector.
* Add user configuration option to enable/disable anomaly detection.
* Do not log dimension-specific operations.
Training and prediction operations happen every second for each
dimension. In prep for making this PR easier to run anomaly detection
for many charts & dimensions, I've removed logs that would cause log
flooding.
* Reset dimensions' bit counter when not above anomaly rate threshold.
* Update the default config options with real values.
With this patch the default configuration options will match the ones
we want our users to use by default.
* Update conditions for creating new ML dimensions.
1. Skip dimensions with update_every != 1,
2. Skip dimensions that come from the ML charts.
With this filtering in place, any configuration value for the
relevant simple_pattern expressions will work correctly.
* Teach buildinfo{,json} about the ML feature.
* Set --enable-ml by default in the configuration options.
This patch is only meant for testing the building of the ML functionality
on Github. It will be reverted once tests pass successfully.
* Minor build system fixes.
- Add path to json header
- Enable C++ linker when ML functionality is enabled
- Rename ml/ml-dummy.cc to ml/ml-dummy.c
* Revert "Set --enable-ml by default in the configuration options."
This reverts commit 28206952a59a577675c86194f2590ec63b60506c.
We pass all Github checks when building the ML functionality, except for
those that run on CentOS 7 due to not having a C++11 toolchain.
* Check for missing dlib and nlohmann files.
We simply check the single-source files upon which our build system
depends. If they are missing, an error message notifies the user
about missing git submodules which are required for the ML
functionality.
* Allow users to specify the maximum number of KMeans iterations.
* Use dlib v19.10
v19.22 broke compatibility with CentOS 7's g++. Development of the
anomaly detection used v19.10, which is the version used by most Debian and
Ubuntu distribution versions that are not past EOL.
No observable performance improvements/regressions specific to the K-Means
algorithm occur between the two versions.
* Detect and use the -std=c++11 flag when building anomaly detection.
This patch automatically adds the -std=c++11 when building netdata
with the ML functionality, if it's supported by the user's toolchain.
With this change we are able to build the agent correctly on CentOS 7.
* Restructure configuration options.
- update default values,
- clamp values to min/max defaults,
- validate and identify conflicting values.
* Add update_every configuration option.
Considerring that the MVP does not support per host configuration
options, the update_every option will be used to filter hosts to train.
With this change anomaly detection will be supported on:
- Single nodes with update_every != 1, and
- Children nodes with a common update_every value that might differ from
the value of the parent node.
* Reorganize anomaly detection charts.
This follows Andrew's suggestion to have four charts to show the number
of anomalous/normal dimensions, the anomaly rate, the detector's window
length, and the events that occur in the prediction step.
Context and family values, along with the necessary information in the
dashboard_info.js file, will be updated in a follow-up commit.
* Do not dump anomaly event info in logs.
* Automatically handle low "train every secs" configuration values.
If a user specifies a very low value for the "train every secs", then
it is possible that the time it takes to train a dimension is higher
than the its allotted time.
In that case, we want the training thread to:
- Reduce it's CPU usage per second, and
- Allow the prediction thread to proceed.
We achieve this by limiting the training time of a single dimension to
be equal to half the time allotted to it. This means, that the training
thread will never consume more than 50% of a single core.
* Automatically detect if ML functionality should be enabled.
With these changes, we enable ML if:
- The user has not explicitly specified --disable-ml, and
- Git submodules have been checked out properly, and
- The toolchain supports C++11.
If the user has explicitly specified --enable-ml, the build fails if
git submodules are missing, or the toolchain does not support C++11.
* Disable anomaly detection by default.
* Do not update charts in locked region.
* Cleanup code reading configuration options.
* Enable C++ linker when building ML.
* Disable ML functionality for CMake builds.
* Skip LGTM for dlib and nlohmann libraries.
* Do not build ML if libuuid is missing.
* Fix dlib path in LGTM's yaml config file.
* Add chart to track duration of prediction step.
* Add chart to track duration of training step.
* Limit the number dimensions in an anomaly event.
This will ensure our JSON results won't grow without any limit. The
default ML configuration options, train approximately ~1700 dimensions
in a newly-installed Netdata agent. The hard-limit is set to 2000
dimensions which:
- Is well above the default number of dimensions we train,
- If it is ever reached it means that the user had accidentaly a
very low anomaly rate threshold, and
- Considering that we sort the result by anomaly score, the cutoff
dimensions will be the less anomalous, ie. the least important to
investigate.
* Add information about the ML charts.
* Update family value in ML charts.
This fix will allow us to show the individual charts in the RHS Anomaly
Detection submenu.
* Rename chart type
s/anomalydetection/anomaly_detection/g
* Expose ML feat in /info endpoint.
* Export ML config through /info endpoint.
* Fix CentOS 7 build.
* Reduce the critical region of a host's lock.
Before this change, each host had a single, dedicated lock to protect
its map of dimensions from adding/deleting new dimensions while training
and detecting anomalies. This was problematic because training of a
single dimension can take several seconds in nodes that are under heavy
load.
After this change, the host's lock protects only the insertion/deletion
of new dimensions, and the prediction step. For the training of dimensions
we use a dedicated lock per dimension, which is responsible for protecting
the dimension from deletion while training.
Prediction is fast enough, even on slow machines or under heavy load,
which allows us to use the host's main lock and avoid increasing the
complexity of our implementation in the anomaly detector.
* Improve the way we are tracking anomaly detector's performance.
This change allows us to:
- track the total training time per update_every period,
- track the maximum training time of a single dimension per
update_every period, and
- export the current number of total, anomalous, normal dimensions
to the /info endpoint.
Also, now that we use dedicated locks per dimensions, we can train under
heavy load continuously without having to sleep in order to yield the
training thread and allow the prediction thread to progress.
* Use samples instead of seconds in ML configuration.
This commit changes the way we are handling input ML configuration
options from the user. Instead of treating values as seconds, we
interpret all inputs as number of update_every periods. This allows
us to enable anomaly detection on hosts that have update_every != 1
second, and still produce a model for training/prediction & detection
that behaves in an expected way.
Tested by running anomaly detection on an agent with update_every = [1,
2, 4] seconds.
* Remove unecessary log message in detection thread
* Move ML configuration to global section.
* Update web/gui/dashboard_info.js
Co-authored-by: Andrew Maguire <andrewm4894@gmail.com>
* Fix typo
Co-authored-by: Andrew Maguire <andrewm4894@gmail.com>
* Rebase.
* Use negative logic for anomaly bit.
* Add info for prediction_stats and training_stats charts.
* Disable ML on PPC64EL.
The CI test fails with -std=c++11 and requires -std=gnu++11 instead.
However, it's not easy to quickly append the required flag to CXXFLAGS.
For the time being, simply disable ML on PPC64EL and if any users
require this functionality we can fix it in the future.
* Add comment on why we disable ML on PPC64EL.
Co-authored-by: Andrew Maguire <andrewm4894@gmail.com>
|
