feat: ec module by c3y1huang · Pull Request #72 · longhorn/spdk

c3y1huang · 2026-05-25T13:47:09Z

Which issue(s) this PR fixes:

Issue longhorn/longhorn#1061

What this PR does / why we need it:

Implement EC module.

Special notes for your reviewer:

None

Additional documentation or context

None

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

…ble fields longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

…queue longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

…gments longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

…d bit clear longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

… counters longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

Copilot

Pull request overview

Copilot reviewed 22 out of 22 changed files in this pull request and generated 3 comments.

Copilot

Pull request overview

Copilot reviewed 22 out of 22 changed files in this pull request and generated 6 comments.

c3y1huang · 2026-05-28T11:52:23Z

+	 * the JSON-RPC response once WIB load completes.
+	 */
+}
+SPDK_RPC_REGISTER("bdev_ec_create", rpc_bdev_ec_create, SPDK_RPC_RUNTIME)


IMO, this is legitimate but low priority for this PR deliverable. Longhorn doesn't read schema/schema.json, impact zero. Can defer to later and improve on request.

Copilot

Pull request overview

Copilot reviewed 22 out of 22 changed files in this pull request and generated 7 comments.

c3y1huang · 2026-05-29T03:57:38Z

+	slot_lba_blocks = ec_bitmap_slot_lba_blocks(ec, ctx->next_copy);
+
+	ec->bitmap_persist_pending = true;
+
+	for (i = 0; i < ec->n; i++) {
+		if (!ec->descs[i] || !ec->bitmap_chans[i] ||
+		    !ec_slot_is_writable(ec, i)) {
+			continue;
+		}
+
+		ctx->writes_in_flight++;
+		rc = spdk_bdev_write(ec->descs[i],
+				     ec->bitmap_chans[i],
+				     ctx->dma_buf,
+				     slot_lba_blocks * ec->bdev.blocklen,
+				     slot_size_bytes,
+				     ec_bitmap_persist_write_cb,
+				     ctx);
+		if (rc != 0) {
+			SPDK_WARNLOG("EC bdev %s: bitmap persist submit failed "
+				     "for slot %u (rc=%d)\n",
+				     ec->bdev.name, i, rc);
+			ctx->writes_in_flight--;


The suggested fix seems under-evaluated (potentially might break the bitmap's single-writer on-disk model). Went with marshalling + home-thread assertion.

I've also under-evaluated. Since Longhorn V2 RAID1 mode supports multi-reactor, marshalling + home-thread assertions are not a good fit, either (1) all write requests would need to be marshalled back to the home-thread, or (2) write requests would get dropped due to home-thread assertions in the multi-reactor case. Neither is safe/scalable approach.

Planning for multi-reactor support.

PoC with a similar approach to the blobstore md_thread pattern.

Copilot

Pull request overview

Copilot reviewed 22 out of 22 changed files in this pull request and generated 7 comments.

+	 * For now the assert lands here: under multi-reactor it aborts at the
+	 * exact spot the follow-up needs to land. Single-reactor is unaffected.
+	 */
+	assert(spdk_get_thread() == ec->home_thread);


+	 * For now the assert lands here: under multi-reactor it aborts at
+	 * the exact spot the follow-up needs to land. Single-reactor is
+	 * unaffected.
+	 */
+	assert(spdk_get_thread() == ec->home_thread);


+	 * For now the assert lands here: under multi-reactor it aborts at
+	 * the exact spot the follow-up needs to land. Single-reactor is
+	 * unaffected.
+	 */
+	assert(spdk_get_thread() == ec->home_thread);


+ * The WIB protects against the RMW write-hole. One dirty bit per
+ * region of EC_WIB_REGION_STRIPES stripes, stored on-disk as two
+ * alternating copies in the last two strips of every parity disk.


+ * Protects against the RMW write-hole. One dirty bit per region of
+ * EC_WIB_REGION_STRIPES stripes, stored on-disk as two alternating
+ * copies in the last two strips of every parity disk. See
+ * bdev_ec_wib.c for the on-disk layout and persist protocol.


+	if (spdk_likely(owner == spdk_get_thread())) {
+		ec_write_into_unmapped_finalize(ec_io);
+	} else {
+		spdk_thread_send_msg(owner, ec_write_into_unmapped_finalize, ec_io);


c3y1huang added 30 commits May 13, 2026 09:07

bdev/nvme: refresh namespace blockcnt during reset reconnect

bc6ba2a

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: add module scaffolding, threading model, and Reed-Solomon ta…

4065e7d

…ble fields longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: bring up create/delete bdev lifecycle

2de9dc6

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: report bdev config and runtime info via JSON dump

a18944f

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: detect base-bdev failure and serve degraded

57651ba

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: hot-swap a failed base bdev

e014d39

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: serve the read path with degraded reconstruction

f84a010

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: serve the full-stripe write path

cbd44b7

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: protect parity with a front-placed Write-Intent Bitmap

42738e1

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: load and merge the WIB across parity disks on startup

d97b456

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: serve sub-stripe writes via RMW under WIB protection

ecb6f68

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: serialize concurrent writers per stripe with stripe-busy claims

48afbb9

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: back-pressure RMW under memory exhaustion

c91cf91

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: defer WRITE_ZEROES to SPDK emulation via is_zero_fill shortcut

0be6c37

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: reconstruct a failed slot via background rebuild walk

199e3e7

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: control the rebuild (start/stop/QoS/progress)

13cf25a

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: scrub dirty regions on startup

5dbe2c6

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: grow capacity in place via bdev_ec_resize

15c5816

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: add the in-band unmapped-bitmap on-disk format

63cb75d

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: persist the unmapped bitmap across disks

67375b5

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: load the unmapped bitmap on startup

daedbb8

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: clear unmapped bits asynchronously via the bit-clear waiter …

1310b29

…queue longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: serve single-stripe UNMAP through the bitmap

e16d24f

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: dispatch unaligned multi-stripe UNMAP via head/inner/tail se…

afd3e4c

…gments longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: synthesize zeros on read of unmapped stripe

ef5c0f2

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: route writes into unmapped stripes via zero-fill and deferre…

5138bb2

…d bit clear longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: expose lifecycle and data-plane JSON-RPC with cumulative I/O…

3a4426a

… counters longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: expose recovery, resize, and status JSON-RPC

9b5cef1

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: add Python CLI and RPC client bindings

cebf45f

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

bdev/ec: refuse delete while rebuild or resize is active

48c5a80

longhorn/longhorn-1061 Signed-off-by: Chin-Ya Huang <chin-ya.huang@suse.com>

c3y1huang force-pushed the 1061-v2-sharding branch 2 times, most recently from 1fc7dfd to 9491893 Compare May 28, 2026 07:43

c3y1huang requested a review from Copilot May 28, 2026 07:48

Copilot started reviewing on behalf of c3y1huang May 28, 2026 07:48 View session

Copilot AI reviewed May 28, 2026

View reviewed changes

Comment thread module/bdev/ec/bdev_ec_rmw.c

Comment thread module/bdev/ec/bdev_ec_rpc.c

Comment thread python/spdk/rpc/bdev.py

c3y1huang force-pushed the 1061-v2-sharding branch 3 times, most recently from c097726 to 82ddb16 Compare May 28, 2026 09:10

c3y1huang requested a review from Copilot May 28, 2026 09:11

Copilot started reviewing on behalf of c3y1huang May 28, 2026 09:11 View session

Copilot AI reviewed May 28, 2026

View reviewed changes

c3y1huang force-pushed the 1061-v2-sharding branch 6 times, most recently from 94ab6a1 to 744e8f0 Compare May 28, 2026 12:09

c3y1huang requested a review from Copilot May 28, 2026 12:09

Copilot started reviewing on behalf of c3y1huang May 28, 2026 12:10 View session

Copilot AI reviewed May 28, 2026

View reviewed changes

c3y1huang force-pushed the 1061-v2-sharding branch 7 times, most recently from c4e0838 to 78af565 Compare May 29, 2026 05:28

c3y1huang requested a review from Copilot May 29, 2026 05:28

Copilot started reviewing on behalf of c3y1huang May 29, 2026 05:29 View session

Copilot AI reviewed May 29, 2026

View reviewed changes

Conversation

c3y1huang commented May 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Which issue(s) this PR fixes:

What this PR does / why we need it:

Special notes for your reviewer:

Additional documentation or context

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

c3y1huang May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

c3y1huang May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

c3y1huang May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

c3y1huang May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

c3y1huang commented May 25, 2026 •

edited

Loading

c3y1huang May 28, 2026 •

edited

Loading

c3y1huang May 29, 2026 •

edited

Loading

c3y1huang May 29, 2026 •

edited

Loading

c3y1huang May 29, 2026 •

edited

Loading