From d87cb07c8e1e4d4717cfcf7bf3d5b7e1cd124bf3 Mon Sep 17 00:00:00 2001
From: Igor Fedotov <ifedotov@suse.com>
Date: Fri, 16 Feb 2018 21:53:12 +0300
Subject: [PATCH] os/bluestore: avoid frequent allocator dump on bluefs
 rebalance failure

Signed-off-by: Igor Fedotov <ifedotov@suse.com>
---
 src/common/legacy_config_opts.h |  3 +++
 src/common/options.cc           |  5 +++++
 src/os/bluestore/BlueStore.cc   | 20 ++++++++++++++++----
 src/os/bluestore/BlueStore.h    |  2 ++
 4 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/src/common/legacy_config_opts.h b/src/common/legacy_config_opts.h
index 298721ef83bde..77828795fe6db 100644
--- a/src/common/legacy_config_opts.h
+++ b/src/common/legacy_config_opts.h
@@ -962,6 +962,9 @@ OPTION(bluestore_bluefs_max_ratio, OPT_FLOAT)  // max fs free / total free
 OPTION(bluestore_bluefs_gift_ratio, OPT_FLOAT) // how much to add at a time
 OPTION(bluestore_bluefs_reclaim_ratio, OPT_FLOAT) // how much to reclaim at a time
 OPTION(bluestore_bluefs_balance_interval, OPT_FLOAT) // how often (sec) to balance free space between bluefs and bluestore
+// how often (sec) to dump allocation failure happened during bluefs rebalance
+OPTION(bluestore_bluefs_balance_failure_dump_interval, OPT_FLOAT)
+
 // If you want to use spdk driver, you need to specify NVMe serial number here
 // with "spdk:" prefix.
 // Users can use 'lspci -vvv -d 8086:0953 | grep "Device Serial Number"' to
diff --git a/src/common/options.cc b/src/common/options.cc
index 628fc57269f6b..889fcb97032cd 100644
--- a/src/common/options.cc
+++ b/src/common/options.cc
@@ -3481,6 +3481,11 @@ std::vector<Option> get_global_options() {
     .set_default(1)
     .set_description("How frequently (in seconds) to balance free space between BlueFS and BlueStore"),
 
+    Option("bluestore_bluefs_balance_failure_dump_interval", Option::TYPE_FLOAT, Option::LEVEL_ADVANCED)
+    .set_default(0)
+    .set_description("How frequently (in seconds) to dump information on "
+      "allocation failure occurred during BlueFS space rebalance"),
+
     Option("bluestore_spdk_mem", Option::TYPE_UINT, Option::LEVEL_DEV)
     .set_default(512)
     .set_description("Amount of dpdk memory size in MB")
diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index bd38f32cb1bd1..88bea601b6212 100644
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -5011,6 +5011,18 @@ int BlueStore::_reconcile_bluefs_freespace()
   return 0;
 }
 
+void BlueStore::_dump_alloc_on_rebalance_failure()
+{
+  auto dump_interval =
+    cct->_conf->bluestore_bluefs_balance_failure_dump_interval;
+  if (dump_interval > 0 &&
+    next_dump_on_bluefs_balance_failure <= ceph_clock_now()) {
+    alloc->dump();
+    next_dump_on_bluefs_balance_failure = ceph_clock_now();
+    next_dump_on_bluefs_balance_failure += dump_interval;
+  }
+}
+
 int BlueStore::_balance_bluefs_freespace(PExtentVector *extents)
 {
   int ret = 0;
@@ -5098,18 +5110,18 @@ int BlueStore::_balance_bluefs_freespace(PExtentVector *extents)
 					0, 0, extents);
 
     if (alloc_len <= 0) {
-      dout(1) << __func__ << " no allocate on 0x" << std::hex << gift
+      dout(0) << __func__ << " no allocate on 0x" << std::hex << gift
               << " min_alloc_size 0x" << min_alloc_size << std::dec << dendl;
       alloc->unreserve(gift);
-      alloc->dump();
+      _dump_alloc_on_rebalance_failure();
       return 0;
     } else if (alloc_len < (int64_t)gift) {
-      dout(1) << __func__ << " insufficient allocate on 0x" << std::hex << gift
+      dout(0) << __func__ << " insufficient allocate on 0x" << std::hex << gift
               << " min_alloc_size 0x" << min_alloc_size 
 	      << " allocated 0x" << alloc_len
 	      << std::dec << dendl;
       alloc->unreserve(gift - alloc_len);
-      alloc->dump();
+      _dump_alloc_on_rebalance_failure();
     }
     for (auto& e : *extents) {
       dout(1) << __func__ << " gifting " << e << " to bluefs" << dendl;
diff --git a/src/os/bluestore/BlueStore.h b/src/os/bluestore/BlueStore.h
index 9bdb12c17116b..edf12f6460782 100644
--- a/src/os/bluestore/BlueStore.h
+++ b/src/os/bluestore/BlueStore.h
@@ -1799,6 +1799,7 @@ private:
   unsigned bluefs_shared_bdev = 0;  ///< which bluefs bdev we are sharing
   bool bluefs_single_shared_device = true;
   utime_t bluefs_last_balance;
+  utime_t next_dump_on_bluefs_balance_failure;
 
   KeyValueDB *db = nullptr;
   BlockDevice *bdev = nullptr;
@@ -1984,6 +1985,7 @@ private:
 
   void _open_statfs();
 
+  void _dump_alloc_on_rebalance_failure();
   int _reconcile_bluefs_freespace();
   int _balance_bluefs_freespace(PExtentVector *extents);
   void _commit_bluefs_freespace(const PExtentVector& extents);
-- 
2.39.5