From: Radosław Zarzyński <rzarzyns@redhat.com>
Date: Wed, 22 May 2024 13:33:23 +0000 (+0200)
Subject: qa: test-erasure-eio.sh honors the EC partial read support
X-Git-Tag: v20.0.0~1565^2~4
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=1308da3a8800430e77b64ce33e9223115388e92f;p=ceph.git

qa: test-erasure-eio.sh honors the EC partial read support

This is supposed to fix:

```
2024-05-15T01:19:55.945 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:243: rados_get_data_bad_size:  rados_get td/test-erasure-
eio pool-jerasure obj-size-81362-1-10 fail
2024-05-15T01:19:55.946 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:104: rados_get:  local dir=td/test-erasure-eio
2024-05-15T01:19:55.946 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:105: rados_get:  local poolname=pool-jerasure
2024-05-15T01:19:55.946 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:106: rados_get:  local objname=obj-size-81362-1-10
2024-05-15T01:19:55.946 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:107: rados_get:  local expect=fail
2024-05-15T01:19:55.946 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:112: rados_get:  '[' fail = fail ']'
2024-05-15T01:19:55.946 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:114: rados_get:  rados --pool pool-jerasure get obj-size-
81362-1-10 td/test-erasure-eio/COPY
2024-05-15T01:19:56.175 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:115: rados_get:  return
2024-05-15T01:19:56.175 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:243: rados_get_data_bad_size:  return 1
2024-05-15T01:19:56.175 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:323: TEST_rados_get_bad_size_shard_1:  return 1
2024-05-15T01:19:56.175 INFO:tasks.workunit.client.0.smithi190.stderr:/home/ubuntu/cephtest/clone.client.0/qa/standalone/erasure-code/test-erasure-eio.sh:41: run:  return 1
```

(https://pulpito.ceph.com/rzarzynski-2024-05-14_22:09:16-rados-wip-osd-ec-partial-reads-distro-default-smithi/7706517/)

The failed scenario was exercising a behavior that got truly
changed by introduction of partial reads. Before, regardless
of read size, OSD was always reading and checking for errors
entire stripe.

In this test first 4 KB has been read from an EC pool with
m=2 k=1 while errors had been injected to shards 1 and 2.
Handling the first 4 KB doesn't really require the damaged
shards but, because of the full-stripe alignment, EIO was
returned. This is not anymore.

Signed-off-by: Radosław Zarzyński <rzarzyns@redhat.com>
---

diff --git a/qa/standalone/erasure-code/test-erasure-eio.sh b/qa/standalone/erasure-code/test-erasure-eio.sh
index 42c538eb9184..4c23b4b4488f 100755
--- a/qa/standalone/erasure-code/test-erasure-eio.sh
+++ b/qa/standalone/erasure-code/test-erasure-eio.sh
@@ -178,9 +178,19 @@ function rados_put_get_data() {
         wait_for_clean || return 1
         # Won't check for eio on get here -- recovery above might have fixed it
     else
-        shard_id=$(expr $shard_id + 1)
-        inject_$inject ec data $poolname $objname $dir $shard_id || return 1
-        rados_get $dir $poolname $objname fail || return 1
+        local another_shard_id=$(expr $shard_id + 1)
+        inject_$inject ec data $poolname $objname $dir $another_shard_id || return 1
+        if [ $shard_id -eq 1 -a $another_shard_id -eq 2 ];
+        then
+            # we're reading 4 kb long object while the stripe size is 8 kb.
+            # as we do partial reads and this request can be satisfied
+            # from the undamaged shard 0, we expect a success.
+            rados_get $dir $poolname $objname || return 1
+        else
+            # both shards 0 and 1 are demaged. there is no way no serve
+            # the requests, regardless of partial reads
+            rados_get $dir $poolname $objname fail || return 1
+        fi
         rm $dir/ORIGINAL
     fi
 
@@ -238,9 +248,19 @@ function rados_get_data_bad_size() {
     rados_get $dir $poolname $objname || return 1
 
     # Leave objname and modify another shard
-    shard_id=$(expr $shard_id + 1)
-    set_size $objname $dir $shard_id $bytes $mode || return 1
-    rados_get $dir $poolname $objname fail || return 1
+    local another_shard_id=$(expr $shard_id + 1)
+    set_size $objname $dir $another_shard_id $bytes $mode || return 1
+    if [ $shard_id -eq 1 -a $another_shard_id -eq 2 ];
+    then
+	# we're reading 4 kb long object while the stripe size is 8 kb.
+	# as we do partial reads and this request can be satisfied
+	# from the undamaged shard 0, we expect a success.
+	rados_get $dir $poolname $objname || return 1
+    else
+	# both shards 0 and 1 are demaged. there is no way no serve
+	# the requests, regardless of partial reads
+	rados_get $dir $poolname $objname fail || return 1
+    fi
     rm $dir/ORIGINAL
 }