There's a lot going on here. First, we were treating "the test is
a Condition" as implying BSI, which it doesn't anymore. Second, the
behavior of conditions was weird and BSI-specific. Third, we had
to propagate these changes and features throughout a bunch of code,
including both the core featurebase code and the DAX replacements/copies
of it, plus the SQL3 layer.
We refactor this so that tests for equality and inequality work for
non-BSI fields, so now if you accidentally use `==` in a Row call
on a non-BSI field, it still works; that's not specific to BSI
fields anymore.
We add a TrackExistence flag to fields, and propagate it through
things like our protobuf code, etcetera, so that we can successfully
create fields. Newly-created fields get this by default, because
we add it unconditionally to them, but the paths that are being
called with existing fields don't add it. So, when we "create"
(really, just load the definition of) a field from something stored
in the schema, we don't add TrackExistence to it, but any path to
creating a new field should.
A time quantum field with NoStandardView will *effectively*
lack TrackExistence.
For sets, mutexes, and time quantums with a standard view, anything
that sets bits will also set a corresponding bit for the record in
a new "existence" view. This allows us to distinguish between an
empty set and a null, and also allows null checks to be constant-time.
When clearing bits, we don't clear existence bits EXCEPT that if
you clear a bit in a mutex, *and the bit actually existed*, we clear
the existence bit. For sets and time quantums, clearing bits never
clears the existence bit.
Deleting records clears the existence bit.
We also add code to the `batch` subpackage to generate suitable
existence field bitmaps and import them. This logic correctly handles
empty sets and nils. The `batch` package does not allow specification
of anything equivalent to clearing a single bit from an existing
record, so we don't have to deal with the mutex complexity in that
case, which is good because it would be impossible.
This requires a number of other subtle changes, such as allowing
new fields to have more than one FieldOption specified for them.
We also drop the handful of implementation bits relating to the
"fullySorted" internal-use-only import flag, which existed only to
support the JSON ingest API, which we've removed.
The most dangerous part of this is that the mutex semantics are
impossible to implement on top of our existing API, because they
require us to know, not how *many* bits we cleared, but which
*specific* bits we cleared. I've implemented this as a new Tx method,
which is almost certainly going to be tech debt one day; if we some
day drop the Import API, we should remove that.
The testing for this is only currently covering the Set/Clear
behavior of PQL, and the Import API. The batch tests haven't been
written yet.
Fields that don't have existence tracking enabled refuse to perform
null/not-null tests. They should also report themselves as having
no null values -- if a record exists, sets in it are considered
empty rather than null.
The SQL3 support requires a number of subtle modifications to both
featurebase and some addon tooling. The essential thing is dropping
the unconditional translation of nil slices to non-nil empty slices
in translateResult, both in the executor and the orchestrator. We
also modify the logic that handles generating results from Extract
calls, to ensure that non-null sets get an empty slice created for
them even if they never have any values assigned.
The expected results for some tests are different now; we expect to
get nil slices, rather than 0-length non-nil slices, for fields which
were never written for a given record. Most tests were not changed.
(In every case, if a test was failing, I actually checked the logic
before changing expected results. This required a lot of tracking down
of edge cases.)
The batch package now rejects as an error attempts to clear single
bits from mutex fields, because so far as I can tell it's simply
impossible to have a roaring import that specifies the correct semantics
there; you can't tell whether to clear an existence bit without
access to the currently-set bits, which the batch API doesn't have.
We already supported the special case of specifying a clear value
of nil for clearing a mutex field; now that is the only allowed
value for a mutex field to have in row.Clears.
We change the logic for fixing up incoming view names (in two places)
to stop assuming that any view in a time field other than "" that does
not have viewStandard as a prefix is a partial time quantum name that
should have "standard_" prepended to it. This allows us to submit
bitmaps for "existence" to time quantum fields and not have them
silently transformed into "standard_existence" because that's what we'd
do with "202203".
We drop the field ClearBits method, which was totally unused.
We drop the sliceDifference function, which was used in a previous
mutex implementation and hasn't been used in ages, and the test
case for it, and the helper function used only by that test case.
* Add (commented out) linters that we should introduce
I went through the available linters and added (commented out) the ones
I think we should work on in the near term. In other words, fix them,
then uncomment them so they are enabled in CI.
* linter: errchkjson
* linter: ineffassign
* linter: gosimple
* linter: errname
* performance counters
* first cut of perf counters and system table fanout and a wire protocol
* significantly refactored prometheus support; removed statsd and exprvar
* removed node_id
* put dax subquery test back
* Change Translator.TranslateFieldIDs method to take a dax.TableKeyer
There are a bunch of other calls to the Translator interface methods
with currently take an `index string`, and those need to be converted to
dax.TableKeyer as well. But I need to review each call, because in at
least one place I noticed one being called with `result.Index` instead
of with the qtbl available. And I don't yet know how those could be
different.
Co-authored-by: Travis Turner <travis@molecula.com>
(cherry picked from commit 7f6ea0e6e5)
- had to make sure we don't snapshot until directive is fully applied
on a computer... otherwise there's races between loading the files and
truncating the write log.
- added a dirty bit to resources and a bool return to incrementing the
write log... don't snapshot if it returns false because that means
there's been no writes. (but make sure you close the storage transaction!)
- added the actually snapshotting routine which just fires every
<timeout> and serially snapshots everything.
- tweaked some logging
- added ability to get all tables in an org/db or literally all. I
think I just needed the "literally all", but it was natural to allow
it to be scoped to org or DB as well.
(cherry picked from commit b8b08bc9eb)
In this commit, the Directive is mocked; it doesn't actually reach out
to a controller.
Limits key translation to only those partitions (per index) specified in
the Directive. Attempting to create or find a key (or ID) for a
partition which is not handled by this node will result in an error;
translation requests are no longer forwarded to other nodes.
Limits import into only those shards specified, per index, by the
Directive. Attempting to import into a shard which is not handled by
this node will result in an error.
Stub out /directive endpoint
The `applyDirective()` method still needs to be implemented.
Update mds references to use the new /mds/types structure
In mds, we moved the shared types to mds/types. FeatureBase needs to
reference those instead.
This also bumps the mds version in go.mod.
Implement the Add/Remove Index part of Holder.ApplyDirective()
This adds functionality to `Holder.ApplyDirective()` which adds or
removes indexes (tables) based on those provided in the Directive. Still
to be implemented here: shards and partitions.
WIP: remove client from Batch
Move Batch into its own package: batch
Also, in order to avoid import loops, this introduces packages:
/batch/types
/client/types
Reorganize the Importer-related code
Moved the Importer interface to package: batch
Move the "pilosa client" implementation of the Importer interface to
package: client
Modify batch.NewBatch to take an Importer (not client)
This commit modifies the batch.NewBatch() function to use a functinal
option on Batch to inject an Importer into the Batch. Prior to this,
NewBatch() took a pointer to a client, which was a little too
restrictive. Now, MDS can implement an Importer which uses information
from MDS to determine to which node(s) the import calls should be directed.
Add client.SetAuthToken() method to satisfy SchemaManager interface
Update ApplyDirective logic to include fields.
This needs more work, but it was enough to get a basic test passing.
Move Transaction type into /types package.
Add interface check on batch.Importer no-op implementation
Updated ApplyDirective to create all currently support Field types
There are still the following TODOs:
- [ ] impolement field options (ex: decimal scale, int min/max, etc).
- [ ] `time` fields
Added support for Decimal.Scale in ApplyDirective
Update mds dependency
Add /health endpoint
Update to use dax (dax/mds) instead of mds.
After moving the mds repository into the dax repository as a
sub-package, this commit changes everything in FeatureBase to use the
dax repo instead of the now abandoned mds repo.
Introduce and implment the WriteLogger interfaces.
This adds both a `WriteLogReader` and `WriteLogWriter` interface. They
are both implemented by the implementation: `fileWriteLogger`. The
`fileWriteLogger` uses the dax/writelogger API to append log messages to
files on disk.
Add WriteLogWriter.ImportRoaring method to interface
This commit adds the `ImportRoaring` method to the `WriteLogWriter`
interface. Still to implement are the `Import` and `ImportValue`
methods.
Reorganize the ApplyDirective code
The primary goal was to cache the incoming Directive on the Holder prior
to applying all of the changes in the directive (i.e. loading data from
the WriteLogger) because applying those changes often validated against
the accepted state of the node.
Implement all of the WriteLogger read/write methods
Implement the HTTP WriteLogger implementation
WIP: Introduce shard.Version. Implement snapshotter.
Add HTTP Snapshotter implementation
This also recofigures server to use the HTTPSnapshotter instead of the
FileSnapshotter.
Implement snapshotter: TableKeys
Implement snapshotter: FieldKeys
Dependency dance
last of the dependency dance
Add support for prototype
This adds the Makefile targets to build the docker container and push it
to ECR.
SQL3 changes which break with dax changes
Missed TODO: implement FieldVersion version to WriteLogger
Address bug causing missing TranslateStores to error
Originally, we tried to limit the TranslateStores which get allocated to
only those for which the node is responsible. This works when adding a
new table. But if a table already exists, there's no logic to start
missing TranslateStores.
This reverts back to the old FeatureBase logic which brutishly allocates
a TranslateStore for every partition, even if one is not needed.
We need to address this by allowing the ApplyDirective logic to
initialize TranslateStores when they don't yet exist.
Move the ImportRoaringShardRequest type to the types package
Since the ImportRoaringShardRequest object is part of the Importer
interface, we need to move it to a non-root (i.e. pilosa) package. All
the other interface types are either concrete types or part of a
sub-package (such as roaring). We do this to prevent an implementer of
the interface from having to import the entire pilosa package and risk
circular imports.
buncha changes to support latest dax stuff
Move dax related types to /dax sub-package
This commit moves all the common "dax" types into the /dax sub-package.
The idea is to ensure that featurebase does not import dax at all.
It's ok if dax imports featurebase.
In the future, we might need to split the dax sub-package (common data
types used by muliple molecula data-plan services) into it's own repo.
Add type: dax.Schema
This isn't currently being used; I started to use is as a replacement
for pilosa_client.Schema, but then deferred that. But we'll need to do
it eventually, so it doesn't hurt to have this type in place.
Export RowIDs.Merge() method for use in orchestrator.
Add CreateSQL method to dax.Table type
The CreateSQL() method will return the "CREATE TABLE" statement required
to create the dax.Table.
Comment out confusing writelogger log message.
We need to revisit this, but for now, this log message is confusing.
Also, rename daxSharder to versionStore.
Remove hard-coded AWS account
Implement more FieldOptions such as Epoch
Some of the FieldOption logic was stubbed out in the dax package. This
commit fills that out more; specifically, it adds the
dax.Field.Options.Epoch parameter.
export stuff needed for TopK in orchestrator
export ValCount stuff to implement Percentile in orchestrator
export more stuff to support less code in orchestrator, shared objs
Port dax repo over to featurebase/dax (run all as sub-services)
This commit does ALOT. Sorry.
It introduces a `featurebase dax` sub-command which can be configured to
run the various dax services as sub-services within the same process, or
individually as the lone service in process.
It also changes all the URL paths to be prefixed with the service name.
So for example, instead of calling localhost:8080/status, you would now
call localhost:8080/featurebase/status.
Also, note that all services provide a /health endpoint to confirm they
are running in process.
Clean up integration tests. Remove PILOSA_ config prefix.
Remove duplicate clients (mistake from porting dax to featurebase)
Rename sub-service "featurebase" to "computer"
In the places where we have hard-coded the sub-service name into a URI
path, I've tried to tag the line with a comment containing:
`// #SERVICEPATHPREFIX`
Update copilot manifest files to reference "computer"
Port dax/README.md from dax repository
Separate (toml) Queryer Config from Injections
We needed to separate the toml config from the configuration required to
inject sub-services into the Queryer. I'm not sure this is the best
solution, but it's *a* solution. So here we are.
Clean up (i.e. remove) the queryer "implementations" package
Remove old test file
Run WriteLogger and Snapshotter as local sub-services.
Prior to this commit, the writelogger and snapshotter services only
worked when run as separate services. This allows them to be run in the
same process as all the other dax services.
There is still some naming issues that we should address, but it's
functional for now.
Clean up (i.e. organize) the intra-service interfaces.
Implement alpha Director for local messages from MDS to Computer
Prior to this commit, messages from MDS to the computer service were
still going over http. This commit introduces an interface
implementation which registers the local computer command, and use that
command's API to directly reference methods used by the Director.
Clean up a few more interface names
Add Queryer OpenAPI document.
Update copilot manifests to reflect latest changes
Add OpenAPI documents for WriteLogger and Snapshotter
Add OpenAPI document for MDS service
Add OpenAPI document for Computer service
Consolidate errors to use fb/errors package.
This commit is a first pass at trying to ensure that all of the DAX code
uses:
"github.com/molecula/featurebase/v3/errors"
This package is a wrapper for "github.com/pkg/errors", so going forward
we want to avoid importing that package.
The only method which isn't backward-compatible is `New()`; the
New() method in the featurebase/errors package takes an errors.Code. If
this becomes a problem, we could change this by reverting New() and then
introducing something like NewCoded(). But for now I think it might
actually discourage someone from just creating a New() error without
thinking about how it should be coded.
Introduce VersionStore interface
Move the existing VersionStore code to the `inmem` package as the
in-memory implementation of the new dax.VersionStore interface.
Introduce NodeService interface
With this, the Controller can maintain a registry of nodes by using this
NodeService interface as opposed to an in-memory map of nodes on the
Controller struct.
This also adds an inmem implementation of the NodeService interface.
Introduce controller.Balancer interface
This moves the existing balancer package to controller/naive package.
The idea is to allow us to add a different Balancer implementation in
the future.
Introduce DirectiveVersion interface
This commit also includes *A LOT* of refactoring to use dax.Worker and
dax.Job types everywhere instead of strings.
Introduce Schemar interface
The previous `Schemar` struct was moved to the `schemar/inmem` package,
and `Schemar` is now an interface implemented by that inmem package.
Remove unused type `nUnit`
Add boltdb implementation of VersionStore interface.
This removed the previous sqlite implementation; we decided not to use
sqlite for now (as a basic, local disk implementation) because it
requires CGO.
--------------------------------------------
No longer applicable:
Add sqlite implementation of VersionStore interface.
This commit implements the VersionStore interface using sqlite. Sqlite
requires CGO, so this may not be something we want to include, but it's
implemented here to get a feel for how an external implementation might
be used; the next step will be to determine how the user configured
FeatureBase to run using sqlite as a backing store for services like
MDS.
Add boltdb implementation of NodeService and DirectiveVersion interfaces.
Add boltdb implementation of naive Balancer interfaces.
This includes the two interfaces defined in `naive/balancer.go`:
- WorkerJobService
- FreeJobService
Add boltdb implementation of Schemar interface.
clean up a linter issue
Thread context.Context through all the interfaces.
Some of the interface implementations are going to use context, so we
need to make that part of the interface. The boltdb implementations, for
example, take a context. This is probably so we can do things like
cancel or timeout operations.
Update interfaces to return error; remove `panic(err)` everywhere.
Down-rev grpc version to 1.38.0
Later versions (after 1.42.0?) cause MustRunCluster.Close() in tests to
deadlock.
This commit also adds an `isComputeNode` feature flag around some of the
write log and shard/partition check functionality so that it doesn't run
under normal conditions (this is excercised by running the sql3 tests
for example).
Add MDS_Persistence test to cover meta data persistence
This adds a basic test which configures the MDS container to use boltdb
as its persistence storage, saved on a docker volume. Then, the mds
container is stopped/replaced, and we confirm that the data stored on
the volume is availble to the new MDS container.
Fix a few things after rebase with sql-experiment branch
The lastest version of sql-experiment contains a fairly significan
refactor of the way query iteration works. This commit adjusts for those
changes.
pull dax IDK changes in to FB IDK (#2177)
* pull dax IDK changes in to FB IDK
* Move docker-related IDK build stuff to featurebase root
Building the docker image required the root level go.mod and vendor
directory. This change moves the make targets to the root level
Makefile, and the Dockerfiles now copy the root level vendor directory
(and everything else in the root for that matter).
* Fix batch- and client-related tests
* InitializePoller on MDS restart/replacement
Prior to this change, if MDS was restarted, its internal poller (which
maintains an in-memory list of nodes to poll) is empty. This is bad,
because it doesn't know about nodes that it should be polling.
This change fixes that. Upon MDS startup, it intializes the poller with
the list of nodes that MDS keeps in persistent storage (currently:
boltdb).
* Add EFS volume to MDS Copilot manifest
This allows us to use MDS's persistent storage (via boltdb) in the
Copilot demo by saving metadata in a boltdb file on EFS.
* Thread logger.Logger through all dax components
* Revert some of the breaking changes from DAX development.
When we first started prototyping DAX, we made changes to the
featurebase core code which would have broken the existing featurebase
functionality. This commit reverts some of those changes. Anywhere that
we need to modify core featurebase functionilty, we put it behind some
kind of feature flag. This flag is typically determined by whether the
running node is a "compute" node (i.e. DAX.COMPUTER.RUN = true).
Co-authored-by: Travis Turner <travis@molecula.com>
add packaging for DAX
need cgo for datagen build
bind to 0.0.0.0, pass GOOS and GOARCH explicitly
not sure if the explicit GOOS/GOARCH is actually necessary...
Get INSERT INTO (aka ingest) working through SQL3
This commit does a few things which I'll try do describe here.
- Introduces a Qctx interface. The existing Qcx is an implementation of
this interface, and can be used exactly how it has been. But this
allows us to abstract away the notion of Qcx in the Queryer (which is
handling SQL3) until we're ready to address that. As an example, the
Qcx has a notion of a featurebase Holder, but that doesn't make sense
when we're at the Queryer layer. For now, the Qctx used in the Queryer
is a no-op.
- Adds a ComputeAPI interface implementation for the Queryer. This is
effectively the Import() and ImportValues() methods used for ingest.
The logic here handles the incoming ImportRequest by first doing any
necessary column and row translation for the entire request, then it
splits the records by shard, and generates a new ImportRequest per
shard with only the shard-appropriate records.
- Changes the mds.Importer to take an MDS interface implementation
(which can be an mds client) instead of an mdsAddress. This allows us
to use a localy MDS implementation rather than assuming we need a
client to make calls over a network.
Add queryer.Importer interface to handle ingest via SQL (#2203)
* Add queryer.Importer interface to handle ingest via SQL
This is meant to support ingest through SQL when the queryer and the
compute services are running in the same process, or when they are on
seperate processes and need to talk via http client.
* remove datagen from RPM
was originally added as a convenience to generate test data, but is
unused and annoying because datagen doesn't easily cross-compile due
to cgo
* add marshalUnmarshal to controller to avoid passing pointers
passing pointers across API boundaries can cause unpredictable things
in local vs remote configurations.
Co-authored-by: Matthew Jaffee <jaffee@pilosa.com>
"fix" a few issues with wrong default partition numbers
these still need to be properly fixed and actually get the correct
data from MDS
go mod tidy
Introduce TableQualifier (OrganizationID/DatabaseID) (#2220)
* add check in ApplyDirective that version is increasing
fix TestAPIDirective to make version always increasing
* fix docker image build and break out dax test in CI
We have to run the DAX integration tests separately as they call out
to Docker, and so it isn't easy to run them in a Docker container as
the other tests do. So we run them directly on the CI runner which has
Docker and Go installed.
We also explicitly exclude these tests from running during the other
tests.
Also my editor was automatically reformatting some comments badly
which is why I added the "data" thing in those two places
* add timeout to poller
* give Poller a default Logger
apparently we can NPE sometimes, seen in CI: https://gitlab.com/molecula/featurebase/-/jobs/3028286364
* bunch of testing fixes, mostly IDK/DAX related
make MDS error if sendDirectives errors, don't just
log. sendDirectives can error if computer nodes disagree about the
validity of a schema (for example), in which case it might need to get
deleted and user notified somehow. very messy, needs more thought.
re-introduce old env prefix to maintain compatibility with master
branch
make self-contained dax container for IDK testing
build IDK images from source (now that all the source is available
since it's in the same repo)
catch errors in DoExtractQuery in idktest.go
fix IDK bug where prefix path was hardcoded in all cases rather than
only when useMDS was true
fix TestBatchTargetMDS... needed to add field options and catch error
when creating table. also needed an _id field
* fix env prefix in tests
* WIP getting tests to pass, wanna see CI
* don't error if we get a zero version directive and we don't have a
directive yet
* cleanup debugging junk
* "fix" future.rename thing, run IDK tests
* Introduce TableQualifier (OrganizationID/DatabaseID)
This commit introduces a lot of new types (in dax/table.go) related to
TableQualifer (which is made up of OrganizationID and DatabaseID), as
well as things like TableID and TableKey.
For the most part, we try to thread a QualifiedTableID through the
entirety of DAX. There are some places (for example in the Balancers,
which are just aware of string keys) which use a string TableKey
(tbl__org__db__tableid).
* Remove some debugging comments
* Add Org/DB support to CLI.
This commit adds support for special commands:
SET
SET ORG acme
SET DB db1
USE db1
* remove ".pulled" from IDK Makefile
I don't think we need it any more as most things can be built
locally. I think it was only there to refresh the FeatureBase images
that were tagged as master, but we don't need to do that any more.
* Change DAX json tags to kebab-case (i.e. hyphenated)
This commit also renames some struct arguments to more accurately
reflect their type: for example, renaming `Table` to `TableKey` when the
type is TableKey.
* Return DAX TableName in SHOW TABLES (instead of Index.Name)
There are cases where SchemaAPI is used to return DAX friendly table
names (as opposed to featurebase index names, which are DAX TableKey).
This is an attempt to do that. With that said, it's not ideal because
anything could call those API methods and expect the other type.
* Fix a bug which wasn't completely dropping a table.
When using boltdb as a backend, DROP TABLE wasn't removing the
reverse-lookup key for the table in boltdb.
* Remove idk/testenv/certs which got accidentally committed.
also update .gitignore to include those.
* Fix IDK ingest tests to be TableQualifier aware.
* Add example Table types to dax/table.com godoc.
* ignore idk.Main fields for flags, upgrade commandeer
* go mod tidy
* Fix DAX integration tests: ingester using wrong ENV VARs
We change from ORGANIZATION_ID to ORG_ID
and from DATABASE_ID to DB_ID
* Clarify things around idk (docker) tests
* Stop running TestKafkaSourceIntegration with t.Parallel()
This test can't be run in parallel as it's currently written. Doing so
allows for interleaving of messages to the same kafka topic between
tests.
I didn't attempt to modify the test so it could be run in parallel. That
could be done, but left for someone more ambitious.
Co-authored-by: Matthew Jaffee <jaffee@pilosa.com>
Require Directive.Version be a non-zero value. (#2227)
Because the directive cached on the holder is not a pointer, its default
version is 0. In order to avoid having to compare against that, we just
require that Directive.Version start at 1.
General, non-invasive code cleanup and comment adjustment.
Move ImportRoaringShardRequest out of the types package
Early on in the DAX development, I moved ImportRoaringShardRequest into
a types package. There must have been some import loop going on, but
since that is not longer the case, it's safe to move this back into the
core featurebase (er... pilosa) package.
Move Transaction struct back into the pilosa package (from types)
Revert some name changes (cli -> client)
Add DAX Handler CloseTimeout
This was implemented in htt_handler.go, but it had been commented out in
the DAX handler. This just uncomments that and finishes the
implementation.
Remove Qcx from queryer.Importer interface
This sets us up to revert the Qctx interface that was initially
introduced to allow us to abstract away the need for a Qcx when calling
the ComputeAPI from a remote service (i.e. the queryer).
Add some go-doc comments and remove unused code.
Move SchemaManager setup from datagen to idk.Main (#2233)
The set for idk.SchemaManager (for dax implementations) was previously
in datagen. This may have been because of some import loop problem
during development, but that's no longer an issue.
The setup for this should be in idk.Main so anything using that can
leverage the MDS-specific SchemaManager setup.
Fix issues around nil TxFactory
First, don't return a nil. Rather return a new *TxFactory (with no
holder).
Second, don't call `f.holder` in the testhook outside of checking if
`f.holder` is nil.
Wrap all bare errors
Make service prefixes constants
Instead of having `"computer"` throughout the code, use instead a
constant: `dax.ServicePrefixComputer`.
MDS skip errors when sending empty directives
also add in the docker-login and ecr-push changes for serverless DAX
Fix the logic in Directive.IsEmpty() (#2236)
Update the cached value for Index.translatePartitions
In the case where a node already knows about an index, but its
assignment of partitions for that index changes (for example, when
another node goes down and the node in question is now responsible for
more partitions than it previously was), then we need to update the
cached value of Index.translatePartitions because that's used in
translation checks.
minor fixes for IDK-related bugs
WIP: tokenize CLI to access cloud
FB CLI cloud support with automatic token refresh
Also adds support for a GET command which allows making HTTP GET
queries to cloud CP which can be handy for debugging stuff. E.g. GET /v2/databases
buncha little fixes working on writelogger stuff
fix writelogger/snapshotter setup bugs
implement writelogging for importRoaringShard
add debug endpoint to MDS
use shard transactional endpoint in MDS datagen
add debugging to API related to writelogger
revert handleroption change
clean up big PR
remove "GET" command from CLI for making arbitrary HTTP request to
cloud control plane (was a messy hack and not that useful)
remove json tags from FB objects where we had to duplicate the object
elsewhere due to import loops and weren't actually json encoding it
unexport handlerOption which was exported to try to avoid doing
certain things if we're in DAX mode, but I didn't end up merging that code.
remove (hopefully) unecessary extra call to api.indexField
fix some formatting, unexport some vars, godoc, etc
oops, fix build failure
Update FeatureBase CLI to support a standard deployment
The standard deployment uses a different endpoint and request payload.
This commit tries to detect is the standard deployment is being used,
and if so, it uses a standard-specific FBQueryer.
It also modifies the auto-detection logic to try standard featurebase
and dax ports in the case where a port was not provided.
MDS API refactor (#2259)
* MDS API refactor
table IDs are exposed but only created server side
also cleaned up dax Makefile
* clean up review feedback
Co-authored-by: Travis Turner <travis@pilosa.com>
* remove TablesByName
* rip out inmem implementations and use boltdb everywhere
* remove inmem balancer, create bolt tempfile by default on startup
* WIP on snapshot table impl and test
* Minor comment and code layout adjustments.
This commit also adds the `Equals` method to `QualifiedTableID` for
equality comparisons. It's no longer safe to compare struct (two structs
might still be equal even if one of the structs doesn't have a `Name`
value.
* Use a unique docker network for each dax test
Ocassionally we would see some test failures due to a network already
existing. This shouldn't happen, but to avoid that, this commit
generates a unique name for each sub test (which gets deleted at the end
of every test).
* Fix one instance of NewQualifiedTableID losing Name
We should probably check the other instances and see if Name is getting
lost.
* simplify unique network stuff and fix api directive tests
* Remove TableIDRequest and TableIDResponse types for /table-id (#2267)
For the mds/table-id http requests, just use dax.QualifiedTableID as
both the request and response types.
* remove lattice from dax, no error on node re-reg, dax docker-compose
* various updates
* WIP: mds-refactor branch review
* no-op on SnapshotTableKeys if table is not keyed
* Makefile helpers
* add doWeCare so controller doesn't fail unnecessarily
* clean up table creation (#2272)
* Strip underscores from TableID stub name
* fix boltdb versionstore tests: generate unique, sorted tables
* fix controller test related to reregistering a node
* JobSet -> generic Set
Co-authored-by: Travis Turner <travis@pilosa.com>
Co-authored-by: Travis Turner <travis@molecula.com>
Cleanup after rebase on master
The latest rebase on master entailed all the client/batch changes as
well as some of the qcx refactoring. It made for a hairy rebase. This
commit fixes some of the tests that were failing after that rebase.
Fix batch/client import loop missed during rebase (#2280)
It's not surprising that `batch` can't import `client`. It was doing
that here (importing an error type from the `client` package). What is
surprising is that it's okay for `batch_test.go` to import `client` even
though `batch_test.go` is an internal test and therefore part of the
`batch` package.
different boltDB's for schemar/controller, explicit balancers
nice helpers for dax docker-compose, make build really fast
build FB binary outside of docker, then create Docker image with its
working dir in an empty subdirectory so it doesn't send a GB of
context to the daemon.
error on unassigned jobs and use client with timeout
fix CR feedback
deregister batch of nodes
also make removal faster via director dial timeout
implement WorkersForJobPrefix so orchestrator doesn't make up shards
also fix some godocs and remove unused method
Run sub-tasks of a Directive concurrently in a worker pool. (#2275)
* Run sub-tasks of a Directive concurrently in a worker pool.
This allows the compute node to concurrently load shapshot and writelog
data concurrently, instead of one keyset/partition/shard at a time.
It introduces a config parameter called `DirectiveWorkerPoolSize`.
* code review cleanup
* Use unique container names in DAX integration tests
We were seeing "container already exists" errors in CI, so just to be
safe, this commit constructs a unique container name for every container
in the DAX integration test run.
Stub in SystemAPI to Queryer (note: will not work if used)
This just makes is so that dax can compile. Actually implementing
system-table functionality for dax will take some planning.
Tlt/dax merge prep (#2282)
* Remove copilot directory
* Remove Dockerfile-datagen-long
* Remove orphaned RegisterNodeRequest
This type is not defined in the dax/mds/http package.
* implement TIMEQUANTUM and TTL in Table.Field type
* Remove the "service" misdirection in queryer/writelogger/snapshotter.
We had originally used an additional layer, er.. package, for a "service".
The main distinction was that the Config differed in that it was
internal, unlike the Config that we need to provide for the top-level
server config (i.e. toml). Having that additional layer just to support
a different Config seemed premature at best. So I'm removing it.
* Remove dax docker containers no longer used in tests
Since we run everything as "featurebase", we don't have multiple
container types anymore.
* Some minor comment updates
* Remove nfpm stuff related to dax
* Fix linter issues
Fix "duplicate" issues raised by sonarcloud.
run docker components of dax integration tests with coverage
trying to get dax integration coverage
add coverate volume mounts throughout dax integration tests
add a lock, tweak dax Makefile, remote flag on query handler
remove some unused code
convert batch tests to use clustertests to get coverage
maybe fix clustertests
more authclustertests fixes, test is failing locally
but also seems to have been silently failing in CI prior to these
changes... let's see if it's still silent
fix some lint to kick CI
just re-running the job wasn't working... strange behavior
remove RetryLogic test and pipe which don't work
RetryLogic test removed due to etcd changes. Seebs thinks we shouldn't
test this here.
Pipe was being ignored since we're no longer using "bash -c" to
execute the command. If we need to generate that output file we'll
either have to reintroduce bash -c and set -o pipefail so that it
actually fails properly, or figure out some other solution.
shooting into the dark...
first cut at bulk node registration
remove unused stuff from batch tests, set coverpkg to ../...
batch registration timeout and fix tests
disable most tests and don't run fb background batch test
debuggin!!!!!!!!!
and then he tried this....
Implement importer (for INSERT INTO) in the Queryer
Prior to this, we we passing a nil value in for the importer to the
planner.NewExecutionPlanner in the Queryer. This meant that INSERT INTO
statements didn't work. Now they should.
It uses the importer that we build for IDK in /idk/mds/importer.go, and
wrapps that with a type that can determine if the provided string
"index" is of the form indexName or TableKey.
turn off debug mode, fix log saving
Run sql3 test definitions in a dax integration test
There are currently 22 tests which are not passing. They are skipped in
the "skips" slice.
WIP, not working, pql queries to tests
Add TableQualifier to PQL query logic in the Queryer
Add more PQL tests to the keyed table
Allow instant node registration if registration-batch-timeout=0
When running dax services in process, we don't want to wait 3s for the
compute node to register; we know it's there because it's in the same
process.
Fixes related to IncludesColumn PQL test.
Tests for ConstRow and FieldValue
cleanup
add UnionRows and Options, better error reporting on bad queries
delete unused schemar client.go, clean up unused in batch test CI
move test timeouts into more reasonable territory
apparently this had already been done, but got merge-stommped at some point
move dax bolt test helpers into dax package
Add computer CheckIn routine (#2296)
* Add computer CheckIn routine
This adds a background routine which sends a "check-in" request to MDS
every <interval>. This is to address the case where the poller has
removed a computer node from the node list (due to a network fault, for
example), but the node is still healthy and becomes available again. In
that case, the node needs to "check-in" to tell MDS it is still there.
MDS will likely send the node a new directive with Method=reset telling
the node to delete all of its data an apply the latest directive.
* Don't send directives to Deregistered (i.e. removed) nodes
We have an issue where we're locking on sendDirective in the
controller, and when the node is unavailable, the send hangs and never
releases the lock. This is a temporary fix for that until we address the
real problem.
Fix .gitlab-ci.yml after rebase
fix some indentation shenanigans
(cherry picked from commit 20a8b5713a)
The anti-entropy feature has never actually worked. We've been
talking about removing it or replacing it for ages, but haven't
had a concrete motivation.
But the anti-entropy interface is the sole user of several components
of the Tx interface, and now that we're trying to replace that
interface, being able to drop those components has some appeal, so
let's remove the one thing that used them, in the hopes that this
will simplify life.
This also lets us drop ForEach and ForEachRange, which were
barely used at all. The one surviving usage (CSV export) can be
handled by using the container iterator we already have, and
making ContainerCallback exported so we can use it to just call
things for every bit.
(cherry picked from commit fff9ddc1f5)
We centralize the creation paths for test indexes, fields,
etcetera so they all have a common path, all using standard
test holders. There's still two versions, one for test.* functions
and one for internal. They do share a TestHolderConfig though.
Large hunks of the related APIs are simplified/streamlined.
* Fragments are always created with a Field and don't need
a workaround in case they don't have it.
* Creation of test fragments, etc., use optional FieldOptions
but don't specify names because they're all using new holders
for each thing created anyway. This dramatically reduces
the complexity of the calls.
* test fragments are created inside test views which are created
inside test fields, etcetera, so everything is using the same
logic; test views aren't bypassing the other layers, they're
creating themselves normally within a field.
* Quite a few things now use the standard runtime/production
logic instead of being custom workarounds; for instance, instead
of `mustOpenMutexFragment` creating a fragment and then creating
a mutex vector for it, we just create a mutex-typed field and
have the normal runtime code do this.
* Similarly, we now use the same field creation logic that production
does, instead of having our own test-only thing that validates
field names directly, so our test that we're validating field names
is actually testing the runtime code. Yay.
* fragSpec goes away. it was a replacement for fragProxy which existed
to solve memory allocation problems but replaced them with interface
overhead problems. Now we just have pointers to things and maintain
valid data structures.
* Many panics are now Fatal or Fatalf calls.
* Some specific bugs fixed, like a cluster which was requested and
then had its first node directly overwritten, which isn't valid with
shared clusters.
* Drop the temp-dir test flag and TempDir variable, we can just use
$TMPDIR.
* Drop a benchmark of "write file to disk" that was purely a benchmark
of file write speed, not a benchmark of rendering the data that needs
to be written.
* Drop the unused "flags" parameter to fragment creation, which was
only used back when we changed the BSI format.
* Use holder.Txf() rather than index.Txf(). The TxFactory has to be
holder-level anyway, referring to it via the index is misleading.
* Test holders automatically close themselves and delete themselves,
we remove various other things that thought they were responsible
for deleting themselves.
A few view functions were taking a Tx, which had to be shard-specific,
but that's sort of awkward -- the view is inherently not shard-specific,
so it should be handling sharding internally.
There were also a couple of remaining obsolete checks for whether a
Tx was nil, at least two of which were in contexts where it absolutely
can't be. Remove all of them, and also the function itself.
* removes unused filesize function
* removes ioutil usage
* updates ioutil.ReadAll to io.ReadAll
* updates ioutil.TempFile to os.CreateTemp
* updates ioutil.TempDir to os.MkdirTemp
* updates ioutil.ReadAll to os.ReadAll
* update ioutil.WriteFile to os.WriteFile
* updates ioutil.Discard to io.Discard
* updates ioutil.ReadDir to os.ReadDir where applicable
* removes unused code in idk
* creates type to use for context value keys
* replaces assert.Nil with assert.NoError for error checks
The new roaring iterator used for the remap and importroaring
things could also be used for unmarshalling roaring streams,
and it's a slightly simpler design that doesn't need two passes
through the data. This patch cleans that up a bit, makes it work
better with ops logs, and uses that instead. It appears to
noticably but not immensely reduce the time imports take, but it
also gets us back down to one thing parsing roaring formats.
There are a couple of subtle changes to errors we were testing
for in various tests, and one of the fuzz tests goes away because
it was actually itself an erroneous error message -- it was reporting
the header of a roaring file as an invalid op because the op log
reader was running on the header for roaring files with zero
containers. Oops.
Annoyingly, this is actually the only place we can make
a read-only lock, because the row() call might write to
the row cache, so it needs the write lock. We might be
able to fix that later, though.
Addressing various lint.
incrementOpN no longer returns errors, because it no longer waits for
the snapshot, so checking those errors is unnecessary.
Several fields in a common embedded structure were "unused" according
to a naive checker.
Other tiny style things, and one actual unchecked error. Yay linters!
If you don't hold the fragment lock when computing rows, it's
pretty reasonable for other stuff to be able to modify it -- which
could invalidate or race the enumeration.
Some calls to f.rows were being made with the lock held, others
weren't, so we introduce `f.unprotectedRows` which has the obvious
semantics. (Without which this looked great except that several
of the tests deadlocked.)
As the size of a fragment grows, the cost of snapshots
increases; with a large fragment getting a lot of large writes,
every write will trigger a snapshot, while any other writes have
to wait for that snapshot before they, too, can trigger a snapshot.
To address this, we introduce a background queue of snapshots.
In general, operations which were omitting their ops log writes
and just snapshotting no longer do; they emit an ops log. This does
mean that, in some cases, the ops log is written and then a snapshot
takes place essentially immediately, which costs us some performance.
However, that only actually happens under very light load; under
heavier load, there's generally going to be multiple writes coalesced
into each snapshot, and the ops log writes for them will be much
cheaper than a full snapshot.
When we do a snapshot, we may end up with containers which are
mmapped to the old file, and containers which have allocated storage
identical to the contents of the new file. It would be nicer if they
were mapped to it. But unmarshalling the entire file is expensive.
Instead, we remap it. (Or, if we couldn't mmap it, just make sure
the old stuff is no longer using the old storage space before we
munmap it.)
Instead of fancy bitmap ops or ImportPositions, we use the
recently-added ImportRoaringBits operations, which can dump
themselves to op logs much more efficiently, and which are
also usually much more efficient than things like "create a
new bitmap which is a copy of the old one".
This patch replaces a lot of circumstances in which containers
were being copied with circumstances in which they are shared,
using copy-on-write semantics.
To achieve this, we emulate somewhat the design of go's
native `append` function. Operations on a container may optionally
yield a new container. A container can be marked "frozen",
after which no operation should ever write to it in any way;
that applies both to the container itself and the backing store
it refers to, if any. So for instance, instead of:
c.arrayToBitmap()
we now write:
c = c.arrayToBitmap()
Operations which need to modify a container in any way
need to be able to return a new container, which is a modified
copy of the previous container. This applies to operations
like add/remove, but also to things like unmapping memory-mapped
storage, or changing a container's type.
Bitmaps do not support the same copy-on-write semantics,
currently, but "copying" a bitmap and sharing the containers
instead of duplicating them is *much* cheaper than copying
the containers.
Bitmaps do support a .Freeze method, which currently copies
the previous bitmap, making a new one with the same container
pointers, and freezes the individual containers. Use this
if you need a writeable copy of a bitmap -- the resulting
bitmap can safely have its set of containers modified, and
bitmap operators that would want to modify the containers
will use copy-on-write for that.
The primary motivation of this is to reduce the cost of the
row cache used by fragments. As a secondary issue, the row cache
is no longer updated on writes -- that update was actually a
race condition waiting to happen. Rather, writes to a row
invalidate the cache entry for that row. The row cache is
created by creating a new bitmap, and freezing the relevant
containers from the fragment's storage. In the case where
nothing is being written, the row cache grows to contain
bitmaps containing all those containers, but never copies
any containers. If nothing's being read, the row cache is
never created, and the containers are in general not getting
frozen. The only circumstance where copies have to happen is
when things are read (and thus stored in the row cache) and
later modified. In that case, each read freezes objects, and
the first write to a container after it's been frozen will
create a new copy.
We drop the enterprise/b btree implementation, because we
don't really need it anymore -- we now provide that
implementation by default in the open source product anyway.
Along with this, there's a lot of other changes which
improve support for nil containers, as a cheaper representation
for empty containers. Operations which we know will provide
an empty container can always short-circuit and just yield
a nil *Container. Similarly, operations which would provide
a full container can return a single shared full container
object (which is frozen). The higher-level (non type-specific)
container ops are now using that logic to short-circuit
operations for empty and full containers. (For instance,
difference of anything minus an empty container is the
original thing, union of anything and empty is the original
thing, and so on.)
The Containers interface adds "Update" and "UpdateEvery"
methods, based in part on the "Put" interface provided
by the underlying btree implementation; Update performs
a possible update in-place of a container for a given
key, bypassing the need to replicate the search for that
key in the container. UpdateEvery loops through all the
containers.
Containers do not strictly guarantee that they won't
return nil `*Container` objects. However, the container
iterators won't return those -- empty containers aren't
interesting. Some tests are updated to reflect this.
Some of the container internals, like N(), or the isArray()
and related functions, accept nil container pointers. Some,
like Thaw(), do not. For the array(), bitmap(), and runs()
methods, roaringparanoia enables an explicit panic on a nil
container explaining the problem, but the intent is that those
should never be called unless you already know you have the
right kind of container, so by default they don't perform
the extra checks. In most cases, this is already covered
because a nil container is empty, and there's no operation
we can perform that requires us to inspect the contents of
an empty container. This is passing a fair amount of testing,
but the testing may not be comprehensive enough.
The overall impact of this is pretty trivial performance-wise.
In our default roaring/ benchmarks, a few things get a few
percent faster, or slower. The advantage is that, with
read-heavy workloads, the row cache no longer eats up incredible
amounts of memory.
For a smallish test case, pilosa's memory usage (RES in top) after
startup was ~2.5GB. Without this patch, simply reading every
row a few times got memory usage to about 9GB, which seemed
reasonably stable. With this patch, memory usage went to about
3GB. This will be less noticeable in mixed read/write loads,
but it should be consistently significantly lower.
In addition to dropping things from the rowCache on modifications,
we also stopped performing a full count on a modified row when
not using a cache of a kind that would use that count, and don't
repopulate the rowCache regardless. We don't want every write
to imply a corresponding read after it.
There's a lot of room for possible future optimizations in
terms of things like in-place operations, and some of the
row/rowSegment code is a little suspicious to me, but I don't
think it should be *worse* in any cases.
This commit implements BSI with variable bit depth using a
sign magnitudeto indicate whether a value is positive or negative.
This also rearranges the existence bit to be the first bit instead
of the last bit.
So with the switch to a new linter, we get a lot of new warnings,
and the majority of them are harmless probably, but a few might be
real. Variously just use _ to suppress warnings, or report errors.
There's probably things here that deserve better fixes, but we can
always revisit it.
Data loss was occuring after a cluster restart. The issue was during the
unmarshaling of the op log when multiple values had been written to the log. The
lines in question were like "changed = changed || b.DirectAdd(v)" in which the
DirectAdd would only be executed when changed was initially false, once it was
true, it would never be executed again.
fix large write path—there was a bug because we were iterating backwards over
the small write path to fix that bug, but the large write path needs to iterate
forward. There is enough code difference between the two paths that they are now
two separate methods (which are probably easier to read).
close files after using them if global max is passed.
I originally implemented this without the global count—just always closing files
when done with them, and reopening for new writes. This was crazy slow for that
one test that uses mustSetBits in a big loop. I modified the test to use
importRoaring and everything worked better (though much more slowly).
After adding the global counter, I ran the tests with that one test using
mustSetBits again, and the performance was similar to master. After completing
this PR, I ran the tests with the max limit set to 5—they still passed but were
much slower.
