Merge pull request ceph#1400 from ceph/wip-hint-tests

wip-hint-tests

Reviewed-by: Sage Weil <[email protected]>

qa/workunits/rados/test_alloc_hint.sh

@@ -0,0 +1,168 @@
+#!/bin/bash
+
+set -ex
+shopt -s nullglob # fns glob expansion in expect_alloc_hint_eq()
+
+#
+# Helpers
+#
+
+function get_xml_val() {
+    local xml="$1"
+    local tag="$2"
+
+    local regex=".*<${tag}>(.*)</${tag}>.*"
+    if [[ ! "${xml}" =~ ${regex} ]]; then
+        echo "'${xml}' xml doesn't match '${tag}' tag regex" >&2
+        return 2
+    fi
+
+    echo "${BASH_REMATCH[1]}"
+}
+
+function get_conf_val() {
+    set -e
+
+    local entity="$1"
+    local option="$2"
+
+    local val
+    val="$(sudo ceph daemon "${entity}" config get --format=xml "${option}")"
+    val="$(get_xml_val "${val}" "${option}")"
+
+    echo "${val}"
+}
+
+function setup_osd_data() {
+    for (( i = 0 ; i < "${NUM_OSDS}" ; i++ )); do
+        OSD_DATA[i]="$(get_conf_val "osd.$i" "osd_data")"
+    done
+}
+
+function setup_pgid() {
+    local poolname="$1"
+    local objname="$2"
+
+    local pgid
+    pgid="$(ceph osd map "${poolname}" "${objname}" --format=xml)"
+    pgid="$(get_xml_val "${pgid}" "pgid")"
+
+    PGID="${pgid}"
+}
+
+function expect_alloc_hint_eq() {
+    local expected_extsize="$1"
+
+    for (( i = 0 ; i < "${NUM_OSDS}" ; i++ )); do
+        # e.g., .../25.6_head/foo__head_7FC1F406__19
+        #       .../26.bs1_head/bar__head_EFE6384B__1a_ffffffffffffffff_1
+        local fns=(${OSD_DATA[i]}/current/${PGID}*_head/${OBJ}_*)
+        local count="${#fns[@]}"
+        if [ "${count}" -ne 1 ]; then
+            echo "bad fns count: ${count}" >&2
+            return 2
+        fi
+
+        local extsize
+        extsize="$(sudo xfs_io -c extsize "${fns[0]}")"
+        local extsize_regex="^\[(.*)\] ${fns[0]}$"
+        if [[ ! "${extsize}" =~ ${extsize_regex} ]]; then
+            echo "extsize doesn't match extsize_regex: ${extsize}" >&2
+            return 2
+        fi
+        extsize="${BASH_REMATCH[1]}"
+
+        if [ "${extsize}" -ne "${expected_extsize}" ]; then
+            echo "FAIL: alloc_hint: actual ${extsize}, expected ${expected_extsize}" >&2
+            return 1
+        fi
+    done
+}
+
+#
+# Global setup
+#
+
+EC_K="2"
+EC_M="1"
+NUM_OSDS="$((EC_K + EC_M))"
+
+NUM_PG="12"
+NUM_PGP="${NUM_PG}"
+
+LOW_CAP="$(get_conf_val "osd.0" "filestore_max_alloc_hint_size")"
+HIGH_CAP="$((LOW_CAP * 10))" # 10M, assuming 1M default cap
+SMALL_HINT="$((LOW_CAP / 4))" # 256K, assuming 1M default cap
+BIG_HINT="$((LOW_CAP * 6))" # 6M, assuming 1M default cap
+
+setup_osd_data
+
+#
+# ReplicatedBackend tests
+#
+
+POOL="alloc_hint-rep"
+ceph osd pool create "${POOL}" "${NUM_PG}"
+ceph osd pool set "${POOL}" size "${NUM_OSDS}" # just in case
+
+OBJ="foo"
+setup_pgid "${POOL}" "${OBJ}"
+rados -p "${POOL}" create "${OBJ}"
+
+# Empty object, SMALL_HINT - expect SMALL_HINT
+rados -p "${POOL}" set-alloc-hint "${OBJ}" "${SMALL_HINT}" "${SMALL_HINT}"
+expect_alloc_hint_eq "${SMALL_HINT}"
+
+# Try changing to BIG_HINT (1) - expect LOW_CAP (BIG_HINT > LOW_CAP)
+rados -p "${POOL}" set-alloc-hint "${OBJ}" "${BIG_HINT}" "${BIG_HINT}"
+expect_alloc_hint_eq "${LOW_CAP}"
+
+# Bump the cap to HIGH_CAP
+ceph tell 'osd.*' injectargs "--filestore_max_alloc_hint_size ${HIGH_CAP}"
+
+# Try changing to BIG_HINT (2) - expect BIG_HINT (BIG_HINT < HIGH_CAP)
+rados -p "${POOL}" set-alloc-hint "${OBJ}" "${BIG_HINT}" "${BIG_HINT}"
+expect_alloc_hint_eq "${BIG_HINT}"
+
+ceph tell 'osd.*' injectargs "--filestore_max_alloc_hint_size ${LOW_CAP}"
+
+# Populate object with some data
+rados -p "${POOL}" put "${OBJ}" /etc/passwd
+
+# Try changing back to SMALL_HINT - expect BIG_HINT (non-empty object)
+rados -p "${POOL}" set-alloc-hint "${OBJ}" "${SMALL_HINT}" "${SMALL_HINT}"
+expect_alloc_hint_eq "${BIG_HINT}"
+
+OBJ="bar"
+setup_pgid "${POOL}" "${OBJ}"
+
+# Non-existent object, SMALL_HINT - expect SMALL_HINT (object creation)
+rados -p "${POOL}" set-alloc-hint "${OBJ}" "${SMALL_HINT}" "${SMALL_HINT}"
+expect_alloc_hint_eq "${SMALL_HINT}"
+
+ceph osd pool delete "${POOL}" "${POOL}" --yes-i-really-really-mean-it
+
+#
+# ECBackend tests
+#
+
+POOL="alloc_hint-ec"
+RULE="ecruleset"
+ceph osd crush rule create-erasure "${RULE}" erasure-code-k="${EC_K}" erasure-code-m="${EC_M}" erasure-code-ruleset-failure-domain=osd
+ceph osd pool create "${POOL}" "${NUM_PG}" "${NUM_PGP}" erasure crush_ruleset="${RULE}" erasure-code-k="${EC_K}" erasure-code-m="${EC_M}" erasure-code-ruleset-failure-domain=osd
+
+OBJ="baz"
+setup_pgid "${POOL}" "${OBJ}"
+rados -p "${POOL}" create "${OBJ}"
+
+# Empty object, SMALL_HINT - expect scaled-down SMALL_HINT
+rados -p "${POOL}" set-alloc-hint "${OBJ}" "${SMALL_HINT}" "${SMALL_HINT}"
+expect_alloc_hint_eq "$((SMALL_HINT / EC_K))"
+
+ceph osd pool delete "${POOL}" "${POOL}" --yes-i-really-really-mean-it
+
+#
+# Global teardown
+#
+
+echo "OK"

src/tools/rados/rados.cc

@@ -101,6 +101,8 @@ void usage(ostream& out)
 "   setomapheader <obj-name> <val>\n"
 "   tmap-to-omap <obj-name>          convert tmap keys/values to omap\n"
 "   listwatchers <obj-name>          list the watchers of this object\n"
+"   set-alloc-hint <obj-name> <expected-object-size> <expected-write-size>\n"
+"                                    set allocation hint for an object\n"
 "\n"
 "IMPORT AND EXPORT\n"
 "   import [options] <local-directory> <rados-pool>\n"
@@ -2197,6 +2199,27 @@ static int rados_tool_common(const std::map < std::string, std::string > &opts,
     ret = io_ctx.notify(oid, 0, bl);
     if (ret != 0)
       cerr << "error calling notify: " << ret << std::endl;
+  } else if (strcmp(nargs[0], "set-alloc-hint") == 0) {
+    if (!pool_name || nargs.size() < 4)
+      usage_exit();
+    string err;
+    string oid(nargs[1]);
+    uint64_t expected_object_size = strict_strtoll(nargs[2], 10, &err);
+    if (!err.empty()) {
+      cerr << "couldn't parse expected_object_size: " << err << std::endl;
+      usage_exit();
+    }
+    uint64_t expected_write_size = strict_strtoll(nargs[3], 10, &err);
+    if (!err.empty()) {
+      cerr << "couldn't parse expected_write_size: " << err << std::endl;
+      usage_exit();
+    }
+    ret = io_ctx.set_alloc_hint(oid, expected_object_size, expected_write_size);
+    if (ret < 0) {
+      cerr << "error setting alloc-hint " << pool_name << "/" << oid << ": "
+           << strerror_r(-ret, buf, sizeof(buf)) << std::endl;
+      goto out;
+    }
   } else if (strcmp(nargs[0], "load-gen") == 0) {
     if (!pool_name) {
       cerr << "error: must specify pool" << std::endl;