Worker processes may have memory limitations imposed by Systemd. But
/proc/meminfo shows the entire system memory regardless of these limits.
This results in the scheduler believing the worker has the entire system
memory avaliable and the worker being allocated too many tasks.
This change attempts to read cgroup memory limits for the worker
process. It supports cgroups v1 and v2, and compares cgroup limits
against the system memory and returns the most conservative values to
prevent the worker from being allocated too many tasks and potentially
triggering an OOM event.
This has been pinned for a while under suspicion that it was causing
graphsync stalls. However, the libp2p team has been unable to reliably
reproduce the issue and believes that it's likely a graphsync issue.
Libp2p issue: https://github.com/libp2p/go-yamux/issues/61
Syntaxt of selection is located at
https://pkg.go.dev/github.com/ipld/go-ipld-selector-text-lite#SelectorSpecFromPath
Example use, assuming that:
- The root of the deal is a plain dag-pb unixfs directory
- The directory is not sharded
- The user wants to retrieve the first entry in that directory
lotus client retrieve --miner f0XXXXX --datamodel-path-selector 'Links/0/Hash' bafyROOTCID ~/output
For a much more elaborate example see the top of ./itests/deals_partial_retrieval_test.go
This commit removes badger from the deal-making processes, and
moves to a new architecture with the dagstore as the cental
component on the miner-side, and CARv2s on the client-side.
Every deal that has been handed off to the sealing subsystem becomes
a shard in the dagstore. Shards are mounted via the LotusMount, which
teaches the dagstore how to load the related piece when serving
retrievals.
When the miner starts the Lotus for the first time with this patch,
we will perform a one-time migration of all active deals into the
dagstore. This is a lightweight process, and it consists simply
of registering the shards in the dagstore.
Shards are backed by the unsealed copy of the piece. This is currently
a CARv1. However, the dagstore keeps CARv2 indices for all pieces, so
when it's time to acquire a shard to serve a retrieval, the unsealed
CARv1 is joined with its index (safeguarded by the dagstore), to form
a read-only blockstore, thus taking the place of the monolithic
badger.
Data transfers have been adjusted to interface directly with CARv2 files.
On inbound transfers (client retrievals, miner storage deals), we stream
the received data into a CARv2 ReadWrite blockstore. On outbound transfers
(client storage deals, miner retrievals), we serve the data off a CARv2
ReadOnly blockstore.
Client-side imports are managed by the refactored *imports.Manager
component (when not using IPFS integration). Just like it before, we use
the go-filestore library to avoid duplicating the data from the original
file in the resulting UnixFS DAG (concretely the leaves). However, the
target of those imports are what we call "ref-CARv2s": CARv2 files placed
under the `$LOTUS_PATH/imports` directory, containing the intermediate
nodes in full, and the leaves as positional references to the original file
on disk.
Client-side retrievals are placed into CARv2 files in the location:
`$LOTUS_PATH/retrievals`.
A new set of `Dagstore*` JSON-RPC operations and `lotus-miner dagstore`
subcommands have been introduced on the miner-side to inspect and manage
the dagstore.
Despite moving to a CARv2-backed system, the IPFS integration has been
respected, and it continues to be possible to make storage deals with data
held in an IPFS node, and to perform retrievals directly into an IPFS node.
NOTE: because the "staging" and "client" Badger blockstores are no longer
used, existing imports on the client will be rendered useless. On startup,
Lotus will enumerate all imports and print WARN statements on the log for
each import that needs to be reimported. These log lines contain these
messages:
- import lacks carv2 path; import will not work; please reimport
- import has missing/broken carv2; please reimport
At the end, we will print a "sanity check completed" message indicating
the count of imports found, and how many were deemed broken.
Co-authored-by: Aarsh Shah <aarshkshah1992@gmail.com>
Co-authored-by: Dirk McCormick <dirkmdev@gmail.com>
Co-authored-by: Raúl Kripalani <raul@protocol.ai>
Co-authored-by: Dirk McCormick <dirkmdev@gmail.com>
* main: init implement rpc.Discover RPC method
This implement the basic functionality for the method
over HTTP RPC.
Signed-off-by: meows <b5c6@protonmail.com>
* main,go.mod,go.sum: init example with go-openrpc-reflect lib
Signed-off-by: meows <b5c6@protonmail.com>
Conflicts:
go.mod
go.sum
* main: make variable name human-friendly
Signed-off-by: meows <b5c6@protonmail.com>
* main,go.mod,go.sum: init impl of go-openrp-reflect printing document
Signed-off-by: meows <b5c6@protonmail.com>
Conflicts:
go.mod
go.sum
* go.mod,go.sum: use go-openrpc-reflect and open-rpc/meta-schema hackforks
This is for development only.
Versions need to be bumped when they're ready for use
as canonical remotes.
Signed-off-by: meows <b5c6@protonmail.com>
* main,openrpc,main: refactor openrpc supporting code to own package
This eliminates code duplication.
Signed-off-by: meows <b5c6@protonmail.com>
* main: add rpc.Discover to openrpc document
Signed-off-by: meows <b5c6@protonmail.com>
* openrpc: fix rpc.discover method name casing
Also fixes casing stuff for the rest of Filecoin.
methods.
Signed-off-by: meows <b5c6@protonmail.com>
* Revert "main: add rpc.Discover to openrpc document"
This reverts commit 116898efb10f33e405ac74acb1aa6daefcd46a62.
* main: fix document creation method name
This fixes an issue caused with the latest reverting
commit.
Signed-off-by: meows <b5c6@protonmail.com>
* main,docgen,openrpc: refactor to share api parsing, etc as docgen exported stuff
Signed-off-by: meows <b5c6@protonmail.com>
Makefile: fix docgen refactoring for makefile use of command
Signed-off-by: meows <b5c6@protonmail.com>
* openrpc: add schema.examples to app reflector
There are quite of few of these already registered
for the docgen command, so it makes sense to use
those!
Signed-off-by: meows <b5c6@protonmail.com>
* openrpc: init method pairing examples
Signed-off-by: meows <b5c6@protonmail.com>
* go.mod,go.sum: bump go.mod to use latest meta-schema and openrpc-reflect versions
Signed-off-by: meows <b5c6@protonmail.com>
* openrpc: init SchemaType mapper function
This function will handle the manual configurations
for app-specific data types w/r/t their json schema
representation.
This is useful for cases where the reflect library
is unable to provide a sufficient representation
automatically.
Provided in this commit is an initial implementation
for the integerD type (assuming number are represented
in the API as hexs), and a commonly used cid.Cid type.
Signed-off-by: meows <b5c6@protonmail.com>
* go.mod,go.sum: tame dependencies by bumping etclabscore/go-openrpc-reflect
This removes a problematic dependency
on github.com/ethereum/go-ethereum, which was
imported as a dependency for a couple github.com/etclabscore/go-openrpc-reflect
tests.
etclabscore/go-openrpc-reflect v0.0.36 has removed this
dependency, so this commit is the result of bumping
that version and then running 'go mod tidy'
This is in response to a review at
https://github.com/filecoin-project/lotus/pull/4711#pullrequestreview-535686205
Date: 2020-11-21 06:52:48-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* main: add 'miner' arg to openrpc gen cmd
This allows the command to EITHER
generate the doc for Full or Miner APIs.
See comment for usage.
Date: 2020-11-21 07:48:05-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* docgen: add missing examples for Miner API
Generating the Miner API OpenRPC doc
(via 'go run ./api/openrpc/cmd miner') caused
the example logic to panic because some types
were missing.
This commit adds those missing types, although
I'm not an expert in the API so I can't
suggest that the example values provided are
ideal or well representative.
Date: 2020-11-21 07:50:21-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* build/openrpc/full.json,build/openrpc/miner.json: add build/openrpc/[full/miner].json docs
These will be used as static documents
provided by the rpc.discover method.
Date: 2020-11-21 07:51:39-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* build: init go-rice openrpc static assets
Date: 2020-11-21 08:23:06-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* main: remove rpc.discover implementation from runtime plugin
Instead of generating the doc on the fly,
we're going to serve a static asset.
Rel https://github.com/filecoin-project/lotus/pull/4711#pullrequestreview-535686205
This removes the runtime implementation from the
RPC server construction.
Date: 2020-11-21 08:41:20-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* api,apistruct,common: add Discover(ctx) method to CommonAPI interface and structs
Date: 2020-11-21 08:41:56-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* main: use rpc server method aliasing for rpc.discover
This depends on a currently-forked change at
filecoin-project/go-jsonrpc 8350f9463ee451b187d35c492e32f1b999e80210
which establishes this new method RPCServer.AliasMethod.
This solves the problem that the OpenRPC
spec says that the document should be served
at the system extension-prefixed endpoing
rpc.discover (not Filecoin.Discover).
In fact, the document will be available at BOTH
endpoints, but that duplicity is harmless.
Date: 2020-11-21 09:18:26-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* api,apistruct,build,common: rpc.discover: return json object instead of string
Instead of casting the JSON asset from bytes to string,
unmarshal it to a map[string]interface{} so the
server will provide it as a JSON object.
Date: 2020-11-21 09:27:11-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* Makefile: merge resolve: docsgen command path
Date: 2020-11-22 07:19:36-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* apistruct,main,docgen,openrpc: merge resolve: fix func exporteds, signatures
Date: 2020-11-22 07:31:03-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* go.mod,go.sum: 'get get' auto-bumps version
Date: 2020-11-22 07:31:44-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* Makefile,docgen,main,build/openrpc: refactor openrpc documentation generation
This creates Makefile command docsgen-openrpc-json,
and refactors the docsgen command to generate both
the markdown and openrpc json documents, redirecting
the output of the openrpc json documentation to
the build/openrpc/ directory, where those json
files will be compiled as static assets via go-rice
boxes.
The api/openrpc/cmd now uses usage argumentation
congruent to that of the docgen command (switching
on API context).
Date: 2020-11-22 08:01:18-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* main,docgen_openrpc: rename api/openrpc -> api/docgen-openrpc
Renames the package as well.
This is intended to parallel the
existing docgen package and command
namespacing.
Date: 2020-11-22 10:34:46-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* api,apistruct,docgen,build,build/openrpc: use typed Discover response
Instead of using a map[string]interface{}, use
a typed response for the Discover method implementation.
This avoids having to set a docgen Example for
the generic map[string]interface{} (as an openrpc document)
which both pollutes the generic type and lacks
useful information for the Discover method example.
Date: 2020-11-22 08:31:16-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* apistruct,build,main,impl: implement Discover method for Worker and StorageMiner APIs
Methods return static compiled assets respective
to the APIs.
Date: 2020-11-22 08:57:18-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* docgen_openrpc,build/openrpc: remove timestamping from openrpc doc info
This should allow openrpc docs generated at different
times to be equal. This is important because the CI
(Circle) runs the docgen command and tests that
the output and the source are unchanged (via git diff).
Date: 2020-11-22 10:47:07-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* main,docgen_openrpc,main,build: fix lint issues
Fixes goimports, staticcheck, golint issues.
Date: 2020-11-22 11:06:46-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* docgenopenrpc: fix: don't use an underscore in package name (golint)
Date: 2020-11-22 11:07:53-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* go.sum: fix: mod-tidy-check (run 'go mod tidy')
Date: 2020-11-22 11:09:48-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* go.mod,go.sum: bump filecoin-project/go-jsonrpc dep to latest
This version includes the necessary RPCServer.AliasMethod
method.
Date: 2020-11-23 12:16:15-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* Makefile,main,build,build/openrpc: init gzipped openrpc static docs
Date: 2020-11-24 06:15:06-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* build: refactor gzip reading
Date: 2020-11-24 06:18:34-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* build: add basic test for openrpc doc from static assets
Date: 2020-11-24 06:30:23-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* build: handle reader Close error
This keeps the errcheck linter happy.
Date: 2020-11-24 06:33:14-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* go.sum: run 'go mod tidy'
Date: 2020-11-24 06:36:07-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* go.mod,go.sum: go mod tidy
Tidying up after resolving the merge conflicts
with master at go.mod
Date: 2020-11-24 06:40:45-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* go.mod,go.sum: bump filecoin-project/go-jsonrpc to latest
This is a repeat of 76e6fd2, since the latest merge
to master seems to have reverted this.
Date: 2020-11-24 06:42:30-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* docgenopenrpc,build/openrpc: remove method example pairings, improve schema examples
Removing method example pairings since they were
redundant to schema examples and were not
implemented well.
Improved schema examples by using the ExampleValue
method instead of the map lookup.
Made a note in the comment here that this is
not ideal, since we have to make a shortcut assumption
/workaround by using 'unknown' as the method name
and the typea as its own parent.
Luckily these values aren't heavily used by the
method logic.
Date: 2020-11-27 12:57:36-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* docgenopenrpc: use generic number jsonschema for number types
Previously used an integer schema assuming
hex encoding. It appears, based on review some
of the examples, that this may not be the case.
Obvioussly this schema could be more descriptive,
but just shooting for mostly likely to be
not wrong at this point.
Date: 2020-12-15 14:44:37-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* cmd/lotus,go.mod,go.sum: maybe fix straggling merge resolution conflicts
Date: 2021-01-19 12:30:42-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* build/openrpc/full.json.gz,build/openrpc/miner.json.gz,build/openrpc/worker.json.gz: run 'make docsgen'
Date: 2021-01-19 12:33:55-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* api/apistruct,node/impl: (lint) gofmt
Date: 2021-01-19 12:39:48-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* api/docgen: maybe fix parse error: open ./api: no such file or directory
Date: 2021-01-19 12:52:04-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* api/docgen,build/openrpc: maybe fix no such file error and run 'make docsgen'
Date: 2021-01-19 12:55:52-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* api/docgen: return if AST comment/groupdoc parsing encounters any error
This will returns empty comments/docs maps.
This should fix issues like:
https://app.circleci.com/pipelines/github/filecoin-project/lotus/12445/workflows/4ebadce9-a298-4ad1-939b-f19ef4c0a5bf/jobs/107218
where the environment makes file lookups hard or
impossible.
Date: 2021-01-19 13:04:58-06:00
Signed-off-by: meows <b5c6@protonmail.com>
* api: Don't depend on build/
* make: support parallel docsgen
* openrpc gen: Use simple build version
* methodgen
* goimports
Co-authored-by: meows <b5c6@protonmail.com>
This paves the way for better object lifetime management.
Concretely, it makes it possible to:
- have different stores backing chain and state data.
- having the same datastore library, but using different parameters.
- attach different caching layers/policies to each class of data, e.g.
sizing caches differently.
- specifying different retention policies for chain and state data.
This separation is important because:
- access patterns/frequency of chain and state data are different.
- state is derivable from chain, so one could never expunge the chain
store, and only retain state objects reachable from the last finality
in the state store.
Go 1.16 elides dependencies from go.sum when you don't use a package
which has these dependencies.
This means dependencies for the `rice` command where not in our
`go.sum`.
Signed-off-by: Jakub Sztandera <kubuxu@protocol.ai>
Most changes are just tagging deps. However,
* Updates go-ipld-hamt to fix a bug in the new V3 HAMT.
* Updates bitfield to save ~10% for some operations.
This pulls in the improvements introduced in:
- https://github.com/raulk/go-watchdog/releases/tag/v1.0.0
- https://github.com/raulk/go-watchdog/releases/tag/v1.0.1
Lotus tries to initialize the watchdog in the following order of precedence:
1. If a max heap limit has been provided, initialize a heap-driven watchdog.
2. Else, try to initialize a cgroup-driven watchdog.
3. Else, try to initialize a system-driven watchdog.
4. Else, log a warning that the system is flying solo, and return.
This PR also enabled automatic heap profile capture when memory usage
surpasses 90% of the limit. Profiles are written to <LOTUS_HOME>/heapprof.
A single heap profile is captured per episode, with a max of 10 episodes
captured during the lifetime of the process. Episode = instance of usage
climbing above the 90% threshold.
Instead, use proof type from miner actor. This will, in turn, use the upgraded
proof type if/when it's switched at runtime.
TODO: Consider making this some form of config option instead?
This new release:
* Ensures we send window updates on-time to avoid sawtooth bandwidth utilization.
* Limits the number of outgoing pings to 1 (no need to send any more at a time).
* Limits the number of buffered pongs to 32 (there should never be more than 1,
maybe 2, at a time).
* Fast-tracks outgoing pings so they don't get stuck behind large queued writes.
