From 68ce44ccd56b93ed01b2edb02e24c32d4ae3f535 Mon Sep 17 00:00:00 2001
From: Bilal Akhtar <bilal@cockroachlabs.com>
Date: Mon, 9 Jan 2023 19:08:52 -0500
Subject: [PATCH] *: Add objstorage.shared.Storage interface for storing
 sstables

This change adds a objstorage.shared.Storage interface that can be
implemented by blob storage drivers. The objstorage.Provider
coming in #2267 will largely be responsible for storing
state of files' locations (i.e. shared or local), and calling into
Storage as necessary.
---
 objstorage/shared/storage.go | 52 ++++++++++++++++++++++++++++++++++++
 options.go                   | 12 +++++++++
 2 files changed, 64 insertions(+)
 create mode 100644 objstorage/shared/storage.go

diff --git a/objstorage/shared/storage.go b/objstorage/shared/storage.go
new file mode 100644
index 0000000000..6702e7d2fc
--- /dev/null
+++ b/objstorage/shared/storage.go
@@ -0,0 +1,52 @@
+// Copyright 2023 The LevelDB-Go and Pebble Authors. All rights reserved. Use
+// of this source code is governed by a BSD-style license that can be found in
+// the LICENSE file.
+
+package shared
+
+import "io"
+
+// Storage is an interface for a blob storage driver. This is lower-level
+// than an FS-like interface, however FS/File-like abstractions can be built on
+// top of these methods.
+//
+// TODO(bilal): Consider pushing shared file obsoletion as well as path
+// generation behind this interface.
+type Storage interface {
+	io.Closer
+
+	// ReadObjectAt returns a Reader for reading the object at the requested name
+	// and offset.
+	ReadObjectAt(basename string, offset int64) (io.ReadCloser, int64, error)
+
+	// CreateObject returns a writer for the object at the request name. A new
+	// empty object is created if CreateObject is called on an existing object.
+	//
+	// A Writer *must* be closed via either Close, and if closing returns a
+	// non-nil error, that error should be handled or reported to the user -- an
+	// implementation may buffer written data until Close and only then return
+	// an error, or Write may return an opaque io.EOF with the underlying cause
+	// returned by the subsequent Close().
+	CreateObject(basename string) (io.WriteCloser, error)
+
+	// List enumerates files within the supplied prefix, returning a list of
+	// objects within that prefix. If delimiter is non-empty, names which have the
+	// same prefix, prior to the delimiter but after the prefix, are grouped into a
+	// single result which is that prefix. The order that results are returned is
+	// undefined. If a prefix is specified, the prefix is trimmed from the result
+	// list.
+	//
+	// An example would be, if the storage contains objects a, b/4, b/5 and b/6,
+	// these would be the return values:
+	//   List("", "") -> ["a", "b/4", "b/5", "b/6"]
+	//   List("", "/") -> ["a", "b"]
+	//   List("b", "/") -> ["4", "5", "6"]
+	//   List("b", "") -> ["/4", "/5", "/6"]
+	List(prefix, delimiter string) []string
+
+	// Delete removes the named object from the store.
+	Delete(basename string) error
+
+	// Size returns the length of the named object in bytes.
+	Size(basename string) (int64, error)
+}
diff --git a/options.go b/options.go
index c5016b06f3..fa8f24effd 100644
--- a/options.go
+++ b/options.go
@@ -624,6 +624,18 @@ type Options struct {
 		// https://github.com/cockroachdb/pebble/issues/2292 and
 		// https://github.com/cockroachdb/pebble/issues/2266 are closed.
 		IngestSSTablesAsFlushable bool
+
+		// SharedStorage is a second FS-like storage medium that can be shared
+		// between multiple Pebble instances. It is used to store sstables only, and
+		// is managed by objstorage.Provider. Each sstable might only be written to
+		// by one Pebble instance, but other Pebble instances can possibly read the
+		// same files if they have the path to get to them. The pebble instance that
+		// wrote a file should not delete it if other Pebble instances are known to
+		// be reading this file. This FS is expected to have slower read/write
+		// performance than the default FS above.
+		//
+		// TODO(bilal): Uncomment this once it's in use.
+		// SharedStorage shared.Storage
 	}
 
 	// Filters is a map from filter policy name to filter policy. It is used for