[RFC 0/2] migration: Tally pre-copy, downtime and post-copy bytes independently

[RFC 0/2] migration: Tally pre-copy, downtime and post-copy bytes independently

[David Edmondson]

When examining a report of poor migration behaviour, it would often be
useful to understand how much data was transferred in different phases
of the migration process.

For example, if the downtime limit is exceeded, to know how much data
was transferred during the downtime.

RFC because the name "ram_transferred_add" doesn't seem great, and I'm
unsure whether the tests to determine the phase in the second patch
are the most appropriate.

David Edmondson (2):
  migration: Introduce ram_transferred_add()
  migration: Tally pre-copy, downtime and post-copy bytes independently

 migration/migration.c |  3 +++
 migration/ram.c       | 30 +++++++++++++++++++++---------
 migration/ram.h       |  1 +
 monitor/hmp-cmds.c    | 12 ++++++++++++
 qapi/migration.json   |  4 +++-
 5 files changed, 40 insertions(+), 10 deletions(-)

-- 
2.33.0

[RFC 1/2] migration: Introduce ram_transferred_add()

[David Edmondson]

...and use it.

Signed-off-by: David Edmondson <david.edmondson@oracle.com>
---
 migration/ram.c | 23 ++++++++++++++---------
 migration/ram.h |  1 +
 2 files changed, 15 insertions(+), 9 deletions(-)

diff --git a/migration/ram.c b/migration/ram.c
index 57efa67f20..48ef2819f6 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -386,6 +386,11 @@ uint64_t ram_bytes_remaining(void)
 
 MigrationStats ram_counters;
 
+void ram_transferred_add(uint64_t bytes)
+{
+    ram_counters.transferred += bytes;
+}
+
 /* used by the search for pages to send */
 struct PageSearchStatus {
     /* Current block being searched */
@@ -767,7 +772,7 @@ static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
      * RAM_SAVE_FLAG_CONTINUE.
      */
     xbzrle_counters.bytes += bytes_xbzrle - 8;
-    ram_counters.transferred += bytes_xbzrle;
+    ram_transferred_add(bytes_xbzrle);
 
     return 1;
 }
@@ -1198,7 +1203,7 @@ static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
 
     if (len) {
         ram_counters.duplicate++;
-        ram_counters.transferred += len;
+        ram_transferred_add(len);
         return 1;
     }
     return -1;
@@ -1234,7 +1239,7 @@ static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
     }
 
     if (bytes_xmit) {
-        ram_counters.transferred += bytes_xmit;
+        ram_transferred_add(bytes_xmit);
         *pages = 1;
     }
 
@@ -1265,8 +1270,8 @@ static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
 static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
                             uint8_t *buf, bool async)
 {
-    ram_counters.transferred += save_page_header(rs, rs->f, block,
-                                                 offset | RAM_SAVE_FLAG_PAGE);
+    ram_transferred_add(save_page_header(rs, rs->f, block,
+                                         offset | RAM_SAVE_FLAG_PAGE));
     if (async) {
         qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE,
                               migrate_release_ram() &
@@ -1274,7 +1279,7 @@ static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
     } else {
         qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE);
     }
-    ram_counters.transferred += TARGET_PAGE_SIZE;
+    ram_transferred_add(TARGET_PAGE_SIZE);
     ram_counters.normal++;
     return 1;
 }
@@ -1373,7 +1378,7 @@ exit:
 static void
 update_compress_thread_counts(const CompressParam *param, int bytes_xmit)
 {
-    ram_counters.transferred += bytes_xmit;
+    ram_transferred_add(bytes_xmit);
 
     if (param->zero_page) {
         ram_counters.duplicate++;
@@ -2298,7 +2303,7 @@ void acct_update_position(QEMUFile *f, size_t size, bool zero)
         ram_counters.duplicate += pages;
     } else {
         ram_counters.normal += pages;
-        ram_counters.transferred += size;
+        ram_transferred_add(size);
         qemu_update_position(f, size);
     }
 }
@@ -3133,7 +3138,7 @@ out:
         multifd_send_sync_main(rs->f);
         qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
         qemu_fflush(f);
-        ram_counters.transferred += 8;
+        ram_transferred_add(8);
 
         ret = qemu_file_get_error(f);
     }
diff --git a/migration/ram.h b/migration/ram.h
index c515396a9a..a5b2ffdc18 100644
--- a/migration/ram.h
+++ b/migration/ram.h
@@ -51,6 +51,7 @@ int xbzrle_cache_resize(uint64_t new_size, Error **errp);
 uint64_t ram_bytes_remaining(void);
 uint64_t ram_bytes_total(void);
 void mig_throttle_counter_reset(void);
+void ram_transferred_add(uint64_t bytes);
 
 uint64_t ram_pagesize_summary(void);
 int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len);
-- 
2.33.0

[RFC 2/2] migration: Tally pre-copy, downtime and post-copy bytes independently

[David Edmondson]

Provide information on the number of bytes copied in the pre-copy,
downtime and post-copy phases of migration.

Signed-off-by: David Edmondson <david.edmondson@oracle.com>
---
 migration/migration.c |  3 +++
 migration/ram.c       |  7 +++++++
 monitor/hmp-cmds.c    | 12 ++++++++++++
 qapi/migration.json   |  4 +++-
 4 files changed, 25 insertions(+), 1 deletion(-)

diff --git a/migration/migration.c b/migration/migration.c
index 3de11ae921..3950510be7 100644
--- a/migration/migration.c
+++ b/migration/migration.c
@@ -1013,6 +1013,9 @@ static void populate_ram_info(MigrationInfo *info, MigrationState *s)
     info->ram->page_size = page_size;
     info->ram->multifd_bytes = ram_counters.multifd_bytes;
     info->ram->pages_per_second = s->pages_per_second;
+    info->ram->precopy_bytes = ram_counters.precopy_bytes;
+    info->ram->downtime_bytes = ram_counters.downtime_bytes;
+    info->ram->postcopy_bytes = ram_counters.postcopy_bytes;
 
     if (migrate_use_xbzrle()) {
         info->has_xbzrle_cache = true;
diff --git a/migration/ram.c b/migration/ram.c
index 48ef2819f6..6c5c8441fd 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -388,6 +388,13 @@ MigrationStats ram_counters;
 
 void ram_transferred_add(uint64_t bytes)
 {
+    if (runstate_is_running()) {
+        ram_counters.precopy_bytes += bytes;
+    } else if (migration_in_postcopy()) {
+        ram_counters.postcopy_bytes += bytes;
+    } else {
+        ram_counters.downtime_bytes += bytes;
+    }
     ram_counters.transferred += bytes;
 }
 
diff --git a/monitor/hmp-cmds.c b/monitor/hmp-cmds.c
index 9c91bf93e9..6049772178 100644
--- a/monitor/hmp-cmds.c
+++ b/monitor/hmp-cmds.c
@@ -293,6 +293,18 @@ void hmp_info_migrate(Monitor *mon, const QDict *qdict)
             monitor_printf(mon, "postcopy request count: %" PRIu64 "\n",
                            info->ram->postcopy_requests);
         }
+        if (info->ram->precopy_bytes) {
+            monitor_printf(mon, "precopy ram: %" PRIu64 " kbytes\n",
+                           info->ram->precopy_bytes >> 10);
+        }
+        if (info->ram->downtime_bytes) {
+            monitor_printf(mon, "downtime ram: %" PRIu64 " kbytes\n",
+                           info->ram->downtime_bytes >> 10);
+        }
+        if (info->ram->postcopy_bytes) {
+            monitor_printf(mon, "postcopy ram: %" PRIu64 " kbytes\n",
+                           info->ram->postcopy_bytes >> 10);
+        }
     }
 
     if (info->has_disk) {
diff --git a/qapi/migration.json b/qapi/migration.json
index bbfd48cf0b..69ec9dc3a6 100644
--- a/qapi/migration.json
+++ b/qapi/migration.json
@@ -54,7 +54,9 @@
            'normal-bytes': 'int', 'dirty-pages-rate' : 'int',
            'mbps' : 'number', 'dirty-sync-count' : 'int',
            'postcopy-requests' : 'int', 'page-size' : 'int',
-           'multifd-bytes' : 'uint64', 'pages-per-second' : 'uint64' } }
+           'multifd-bytes' : 'uint64', 'pages-per-second' : 'uint64',
+           'precopy-bytes' : 'uint64', 'downtime-bytes' : 'uint64',
+           'postcopy-bytes' : 'uint64' } }
 
 ##
 # @XBZRLECacheStats:
-- 
2.33.0

Re: [RFC 1/2] migration: Introduce ram_transferred_add()

[Philippe Mathieu-Daudé]

On 12/16/21 13:34, David Edmondson wrote:
> ...and use it.
> 
> Signed-off-by: David Edmondson <david.edmondson@oracle.com>
> ---
>  migration/ram.c | 23 ++++++++++++++---------
>  migration/ram.h |  1 +
>  2 files changed, 15 insertions(+), 9 deletions(-)

> diff --git a/migration/ram.h b/migration/ram.h
> index c515396a9a..a5b2ffdc18 100644
> --- a/migration/ram.h
> +++ b/migration/ram.h
> @@ -51,6 +51,7 @@ int xbzrle_cache_resize(uint64_t new_size, Error **errp);
>  uint64_t ram_bytes_remaining(void);
>  uint64_t ram_bytes_total(void);
>  void mig_throttle_counter_reset(void);
> +void ram_transferred_add(uint64_t bytes);

Why make the method public? It seems an internal operation. Do you
plan to use it elsewhere?

Re: [RFC 1/2] migration: Introduce ram_transferred_add()

[David Edmondson]

On Friday, 2021-12-17 at 20:09:12 +01, Philippe Mathieu-Daudé wrote:

> On 12/16/21 13:34, David Edmondson wrote:
>> ...and use it.
>> 
>> Signed-off-by: David Edmondson <david.edmondson@oracle.com>
>> ---
>>  migration/ram.c | 23 ++++++++++++++---------
>>  migration/ram.h |  1 +
>>  2 files changed, 15 insertions(+), 9 deletions(-)
>
>> diff --git a/migration/ram.h b/migration/ram.h
>> index c515396a9a..a5b2ffdc18 100644
>> --- a/migration/ram.h
>> +++ b/migration/ram.h
>> @@ -51,6 +51,7 @@ int xbzrle_cache_resize(uint64_t new_size, Error **errp);
>>  uint64_t ram_bytes_remaining(void);
>>  uint64_t ram_bytes_total(void);
>>  void mig_throttle_counter_reset(void);
>> +void ram_transferred_add(uint64_t bytes);
>
> Why make the method public? It seems an internal operation. Do you
> plan to use it elsewhere?

No such plan. Will fix in v2.

dme.
-- 
We wanna wait, but here we go again.