lotus/markets/dagstore/wrapper.go

437 lines
13 KiB
Go
Raw Normal View History

integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
package dagstore
import (
"context"
"errors"
"fmt"
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
"math"
"os"
"path/filepath"
"sync"
"time"
"github.com/ipfs/go-cid"
ds "github.com/ipfs/go-datastore"
levelds "github.com/ipfs/go-ds-leveldb"
measure "github.com/ipfs/go-ds-measure"
logging "github.com/ipfs/go-log/v2"
2022-06-14 15:00:51 +00:00
carindex "github.com/ipld/go-car/v2/index"
2022-08-25 18:20:41 +00:00
"github.com/libp2p/go-libp2p/core/host"
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
ldbopts "github.com/syndtr/goleveldb/leveldb/opt"
"golang.org/x/xerrors"
"github.com/filecoin-project/dagstore"
"github.com/filecoin-project/dagstore/index"
"github.com/filecoin-project/dagstore/mount"
"github.com/filecoin-project/dagstore/shard"
"github.com/filecoin-project/go-fil-markets/storagemarket"
"github.com/filecoin-project/go-fil-markets/storagemarket/impl/providerstates"
"github.com/filecoin-project/go-fil-markets/stores"
2022-06-14 15:00:51 +00:00
"github.com/filecoin-project/go-statemachine/fsm"
"github.com/filecoin-project/lotus/node/config"
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
)
const (
maxRecoverAttempts = 1
shardRegMarker = ".shard-registration-complete"
)
var log = logging.Logger("dagstore")
type Wrapper struct {
ctx context.Context
cancel context.CancelFunc
backgroundWg sync.WaitGroup
cfg config.DAGStoreConfig
dagst dagstore.Interface
minerAPI MinerAPI
failureCh chan dagstore.ShardResult
gcInterval time.Duration
}
var _ stores.DAGStoreWrapper = (*Wrapper)(nil)
2021-11-18 08:07:12 +00:00
func NewDAGStore(cfg config.DAGStoreConfig, minerApi MinerAPI, h host.Host) (*dagstore.DAGStore, *Wrapper, error) {
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
// construct the DAG Store.
registry := mount.NewRegistry()
if err := registry.Register(lotusScheme, mountTemplate(minerApi)); err != nil {
return nil, nil, xerrors.Errorf("failed to create registry: %w", err)
}
// The dagstore will write Shard failures to the `failureCh` here.
failureCh := make(chan dagstore.ShardResult, 1)
var (
transientsDir = filepath.Join(cfg.RootDir, "transients")
datastoreDir = filepath.Join(cfg.RootDir, "datastore")
indexDir = filepath.Join(cfg.RootDir, "index")
)
dstore, err := newDatastore(datastoreDir)
if err != nil {
return nil, nil, xerrors.Errorf("failed to create dagstore datastore in %s: %w", datastoreDir, err)
}
irepo, err := index.NewFSRepo(indexDir)
if err != nil {
return nil, nil, xerrors.Errorf("failed to initialise dagstore index repo: %w", err)
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
}
2022-02-18 10:00:01 +00:00
topIndex := index.NewInverted(dstore)
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
dcfg := dagstore.Config{
TransientsDir: transientsDir,
IndexRepo: irepo,
Datastore: dstore,
MountRegistry: registry,
FailureCh: failureCh,
2021-11-18 08:07:12 +00:00
TopLevelIndex: topIndex,
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
// not limiting fetches globally, as the Lotus mount does
// conditional throttling.
MaxConcurrentIndex: cfg.MaxConcurrentIndex,
MaxConcurrentReadyFetches: cfg.MaxConcurrentReadyFetches,
RecoverOnStart: dagstore.RecoverOnAcquire,
}
dagst, err := dagstore.NewDAGStore(dcfg)
if err != nil {
return nil, nil, xerrors.Errorf("failed to create DAG store: %w", err)
}
w := &Wrapper{
cfg: cfg,
dagst: dagst,
minerAPI: minerApi,
failureCh: failureCh,
gcInterval: time.Duration(cfg.GCInterval),
}
return dagst, w, nil
}
// newDatastore creates a datastore under the given base directory
// for dagstore metadata.
func newDatastore(dir string) (ds.Batching, error) {
// Create the datastore directory if it doesn't exist yet.
if err := os.MkdirAll(dir, 0755); err != nil {
return nil, xerrors.Errorf("failed to create directory %s for DAG store datastore: %w", dir, err)
}
// Create a new LevelDB datastore
dstore, err := levelds.NewDatastore(dir, &levelds.Options{
Compression: ldbopts.NoCompression,
NoSync: false,
Strict: ldbopts.StrictAll,
ReadOnly: false,
})
if err != nil {
return nil, xerrors.Errorf("failed to open datastore for DAG store: %w", err)
}
// Keep statistics about the datastore
mds := measure.New("measure.", dstore)
return mds, nil
}
func (w *Wrapper) Start(ctx context.Context) error {
w.ctx, w.cancel = context.WithCancel(ctx)
// Run a go-routine to do DagStore GC.
w.backgroundWg.Add(1)
go w.gcLoop()
// Run a go-routine for shard recovery
if dss, ok := w.dagst.(*dagstore.DAGStore); ok {
w.backgroundWg.Add(1)
go dagstore.RecoverImmediately(w.ctx, dss, w.failureCh, maxRecoverAttempts, w.backgroundWg.Done)
}
return w.dagst.Start(ctx)
}
func (w *Wrapper) gcLoop() {
defer w.backgroundWg.Done()
ticker := time.NewTicker(w.gcInterval)
defer ticker.Stop()
for w.ctx.Err() == nil {
select {
// GC the DAG store on every tick
case <-ticker.C:
_, _ = w.dagst.GC(w.ctx)
// Exit when the DAG store wrapper is shutdown
case <-w.ctx.Done():
return
}
}
}
func (w *Wrapper) LoadShard(ctx context.Context, pieceCid cid.Cid) (stores.ClosableBlockstore, error) {
log.Debugf("acquiring shard for piece CID %s", pieceCid)
key := shard.KeyFromCID(pieceCid)
resch := make(chan dagstore.ShardResult, 1)
err := w.dagst.AcquireShard(ctx, key, resch, dagstore.AcquireOpts{})
log.Debugf("sent message to acquire shard for piece CID %s", pieceCid)
if err != nil {
if !errors.Is(err, dagstore.ErrShardUnknown) {
return nil, xerrors.Errorf("failed to schedule acquire shard for piece CID %s: %w", pieceCid, err)
}
// if the DAGStore does not know about the Shard -> register it and then try to acquire it again.
log.Warnw("failed to load shard as shard is not registered, will re-register", "pieceCID", pieceCid)
// The path of a transient file that we can ask the DAG Store to use
// to perform the Indexing rather than fetching it via the Mount if
// we already have a transient file. However, we don't have it here
// and therefore we pass an empty file path.
carPath := ""
if err := stores.RegisterShardSync(ctx, w, pieceCid, carPath, false); err != nil {
return nil, xerrors.Errorf("failed to re-register shard during loading piece CID %s: %w", pieceCid, err)
}
log.Warnw("successfully re-registered shard", "pieceCID", pieceCid)
resch = make(chan dagstore.ShardResult, 1)
if err := w.dagst.AcquireShard(ctx, key, resch, dagstore.AcquireOpts{}); err != nil {
return nil, xerrors.Errorf("failed to acquire Shard for piece CID %s after re-registering: %w", pieceCid, err)
}
}
// TODO: The context is not yet being actively monitored by the DAG store,
// so we need to select against ctx.Done() until the following issue is
// implemented:
// https://github.com/filecoin-project/dagstore/issues/39
var res dagstore.ShardResult
select {
case <-ctx.Done():
return nil, ctx.Err()
case res = <-resch:
if res.Error != nil {
return nil, xerrors.Errorf("failed to acquire shard for piece CID %s: %w", pieceCid, res.Error)
}
}
bs, err := res.Accessor.Blockstore()
if err != nil {
return nil, err
}
log.Debugf("successfully loaded blockstore for piece CID %s", pieceCid)
return &Blockstore{ReadBlockstore: bs, Closer: res.Accessor}, nil
}
func (w *Wrapper) RegisterShard(ctx context.Context, pieceCid cid.Cid, carPath string, eagerInit bool, resch chan dagstore.ShardResult) error {
// Create a lotus mount with the piece CID
key := shard.KeyFromCID(pieceCid)
mt, err := NewLotusMount(pieceCid, w.minerAPI)
if err != nil {
return xerrors.Errorf("failed to create lotus mount for piece CID %s: %w", pieceCid, err)
}
// Register the shard
opts := dagstore.RegisterOpts{
ExistingTransient: carPath,
LazyInitialization: !eagerInit,
}
err = w.dagst.RegisterShard(ctx, key, mt, resch, opts)
if err != nil {
return xerrors.Errorf("failed to schedule register shard for piece CID %s: %w", pieceCid, err)
}
log.Debugf("successfully submitted Register Shard request for piece CID %s with eagerInit=%t", pieceCid, eagerInit)
return nil
}
2022-07-11 09:30:12 +00:00
func (w *Wrapper) DestroyShard(ctx context.Context, pieceCid cid.Cid, resch chan dagstore.ShardResult) error {
key := shard.KeyFromCID(pieceCid)
opts := dagstore.DestroyOpts{}
err := w.dagst.DestroyShard(ctx, key, resch, opts)
if err != nil {
return xerrors.Errorf("failed to schedule destroy shard for piece CID %s: %w", pieceCid, err)
}
log.Debugf("successfully submitted destroy Shard request for piece CID %s", pieceCid)
return nil
}
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
func (w *Wrapper) MigrateDeals(ctx context.Context, deals []storagemarket.MinerDeal) (bool, error) {
log := log.Named("migrator")
// Check if all deals have already been registered as shards
isComplete, err := w.registrationComplete()
if err != nil {
return false, xerrors.Errorf("failed to get dagstore migration status: %w", err)
}
if isComplete {
// All deals have been registered as shards, bail out
log.Info("no shard migration necessary; already marked complete")
return false, nil
}
log.Infow("registering shards for all active deals in sealing subsystem", "count", len(deals))
inSealingSubsystem := make(map[fsm.StateKey]struct{}, len(providerstates.StatesKnownBySealingSubsystem))
for _, s := range providerstates.StatesKnownBySealingSubsystem {
inSealingSubsystem[s] = struct{}{}
}
// channel where results will be received, and channel where the total
// number of registered shards will be sent.
resch := make(chan dagstore.ShardResult, 32)
totalCh := make(chan int)
doneCh := make(chan struct{})
// Start making progress consuming results. We won't know how many to
// actually consume until we register all shards.
//
// If there are any problems registering shards, just log an error
go func() {
defer close(doneCh)
var total = math.MaxInt64
var res dagstore.ShardResult
for rcvd := 0; rcvd < total; {
select {
case total = <-totalCh:
// we now know the total number of registered shards
// nullify so that we no longer consume from it after closed.
close(totalCh)
totalCh = nil
case res = <-resch:
rcvd++
if res.Error == nil {
log.Infow("async shard registration completed successfully", "shard_key", res.Key)
} else {
log.Warnw("async shard registration failed", "shard_key", res.Key, "error", res.Error)
}
}
}
}()
// Filter for deals that are handed off.
//
// If the deal has not yet been handed off to the sealing subsystem, we
// don't need to call RegisterShard in this migration; RegisterShard will
// be called in the new code once the deal reaches the state where it's
// handed off to the sealing subsystem.
var registered int
for _, deal := range deals {
pieceCid := deal.Proposal.PieceCID
// enrich log statements in this iteration with deal ID and piece CID.
log := log.With("deal_id", deal.DealID, "piece_cid", pieceCid)
// Filter for deals that have been handed off to the sealing subsystem
if _, ok := inSealingSubsystem[deal.State]; !ok {
log.Infow("deal not ready; skipping")
continue
}
log.Infow("registering deal in dagstore with lazy init")
// Register the deal as a shard with the DAG store with lazy initialization.
// The index will be populated the first time the deal is retrieved, or
// through the bulk initialization script.
err = w.RegisterShard(ctx, pieceCid, "", false, resch)
if err != nil {
log.Warnw("failed to register shard", "error", err)
continue
}
registered++
}
log.Infow("finished registering all shards", "total", registered)
totalCh <- registered
<-doneCh
log.Infow("confirmed registration of all shards")
// Completed registering all shards, so mark the migration as complete
err = w.markRegistrationComplete()
if err != nil {
log.Errorf("failed to mark shards as registered: %s", err)
} else {
log.Info("successfully marked migration as complete")
}
log.Infow("dagstore migration complete")
return true, nil
}
// Check for the existence of a "marker" file indicating that the migration
// has completed
func (w *Wrapper) registrationComplete() (bool, error) {
path := filepath.Join(w.cfg.RootDir, shardRegMarker)
_, err := os.Stat(path)
if os.IsNotExist(err) {
return false, nil
}
if err != nil {
return false, err
}
return true, nil
}
// Create a "marker" file indicating that the migration has completed
func (w *Wrapper) markRegistrationComplete() error {
path := filepath.Join(w.cfg.RootDir, shardRegMarker)
file, err := os.Create(path)
if err != nil {
return err
}
return file.Close()
}
// Get all the pieces that contain a block
func (w *Wrapper) GetPiecesContainingBlock(blockCID cid.Cid) ([]cid.Cid, error) {
// Pieces are stored as "shards" in the DAG store
2022-02-18 10:00:01 +00:00
shardKeys, err := w.dagst.ShardsContainingMultihash(w.ctx, blockCID.Hash())
if err != nil {
return nil, xerrors.Errorf("getting pieces containing block %s: %w", blockCID, err)
}
// Convert from shard key to cid
pieceCids := make([]cid.Cid, 0, len(shardKeys))
for _, k := range shardKeys {
c, err := cid.Parse(k.String())
if err != nil {
prefix := fmt.Sprintf("getting pieces containing block %s:", blockCID)
return nil, xerrors.Errorf("%s converting shard key %s to piece cid: %w", prefix, k, err)
}
pieceCids = append(pieceCids, c)
}
return pieceCids, nil
}
2021-11-10 16:28:23 +00:00
func (w *Wrapper) GetIterableIndexForPiece(pieceCid cid.Cid) (carindex.IterableIndex, error) {
return w.dagst.GetIterableIndex(shard.KeyFromCID(pieceCid))
}
integrate DAG store and CARv2 in deal-making (#6671) 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>
2021-08-16 22:34:32 +00:00
func (w *Wrapper) Close() error {
// Cancel the context
w.cancel()
// Close the DAG store
log.Info("will close the dagstore")
if err := w.dagst.Close(); err != nil {
return xerrors.Errorf("failed to close dagstore: %w", err)
}
log.Info("dagstore closed")
// Wait for the background go routine to exit
log.Info("waiting for dagstore background wrapper goroutines to exit")
w.backgroundWg.Wait()
log.Info("exited dagstore background wrapper goroutines")
return nil
}