Compare commits

..

1 Commits

Author SHA1 Message Date
Chris Kirby
253f049251 Don't overrun the block budget in server_log_merge_free_work().
This fixes a potential fence post failure like the following:

error: 1 holders exceeded alloc budget av: bef 7407 now 7392, fr: bef 8185 now 7672

The code is only accounting for the freed btree blocks, not the dirtying of
other items. So it's possible to be at exactly (COMMIT_HOLD_ALLOC_BUDGET / 2),
dirty some log btree blocks, loop again, then consume another
(COMMIT_HOLD_ALLOC_BUDGET / 2) and blow past the total budget.

In this example, we went over by 13 blocks.

By only consuming up to 1/8 of the budget on each loop, and committing when we
have consumed 3/4 of the budget, we can avoid the fence post condition.

Signed-off-by: Chris Kirby <ckirby@versity.com>
2025-05-28 13:48:22 -05:00
2 changed files with 2 additions and 23 deletions

View File

@@ -1,27 +1,6 @@
Versity ScoutFS Release Notes
=============================
---
v1.25
\
*Jun 3, 2025*
Fix a bug that could cause indefinite retries of failed client commits.
Under specific error conditions the client and server's understanding of
the current client commit could get out of sync. The client would retry
commits indefinitely that could never succeed. This manifested as
infinite "critical transaction commit failure" messages in the kernel
log on the client and matching "error <nr> committing client logs" on
the server.
Fix a bug in a specific case of server error handling that could result
in sending references to unwritten blocks to the client. The client
would try to read blocks that hadn't been written and return spurious
errors. This was seen under low free space conditions on the server and
resulted in error messages with error code 116 (The errno enum for
ESTALE, the client's indication that it couldn't read the blocks that it
expected.)
---
v1.24
\

View File

@@ -2531,7 +2531,7 @@ static void server_log_merge_free_work(struct work_struct *work)
ret = scoutfs_btree_free_blocks(sb, &server->alloc,
&server->wri, &fr.key,
&fr.root, COMMIT_HOLD_ALLOC_BUDGET / 2);
&fr.root, COMMIT_HOLD_ALLOC_BUDGET / 8);
if (ret < 0) {
err_str = "freeing log btree";
break;
@@ -2550,7 +2550,7 @@ static void server_log_merge_free_work(struct work_struct *work)
/* freed blocks are in allocator, we *have* to update fr */
BUG_ON(ret < 0);
if (server_hold_alloc_used_since(sb, &hold) >= COMMIT_HOLD_ALLOC_BUDGET / 2) {
if (server_hold_alloc_used_since(sb, &hold) >= (COMMIT_HOLD_ALLOC_BUDGET * 3) / 4) {
mutex_unlock(&server->logs_mutex);
ret = server_apply_commit(sb, &hold, ret);
commit = false;