Invoke-SqlAssessment - IndexFragmentation rule doesn't take partitions into consideration #1260
Comments
|
Seems like something got updated in a newer ruleset - but it still doesn't work for partitioned tables.
|
|
Ok, me and @ClaudioESSilva spent some time this afternoon looking into how this could be resolved and it looks like the following two changes would resolve this issue and allow the SQL Assessment to be more informative and correct when running against environments with partitioned tables. Changing the probe query and the message for the MessageChange to include a spot for partitionNumber (this only contains text if the table is partitioned)
"message": "Remove fragmentation of @{IndexFullName} index@{partitionNumber} Current fragmentation level is @{fragmentation:#0.##}%",Probe QueriesThere are two queries depending on the version of the SQL Instance This is the query for Change to group by partition number and if it is a partitioned table, add the partition number information so we can use it in the message SELECT CASE i.[type] WHEN 0 THEN CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], ' (HEAP)') ELSE CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], '.', i.[name]) END AS IndexFullName , stat.fragmentation, stat.page_count, i.[type] AS IndexType, CASE WHEN p.index_id IS NULL THEN '.' ELSE ' (Partition Number: ' + CAST(stat.partition_number AS varchar(5)) + ').' END AS partitionNumber FROM sys.indexes AS i WITH (NOLOCK) INNER JOIN sys.tables AS t WITH (NOLOCK) ON t.[object_id] = i.[object_id] AND t.is_ms_shipped = 0 INNER JOIN ( SELECT [object_id], index_id, SUM(ps.avg_fragmentation_in_percent) AS fragmentation, SUM(ps.page_count) AS page_count , ps.partition_number FROM sys.dm_db_index_physical_stats(DB_ID(), NULL, NULL, NULL, NULL) ps WHERE ps.index_level = 0 AND ps.alloc_unit_type_desc = 'IN_ROW_DATA' GROUP BY [object_id], index_id , ps.partition_number ) stat ON stat.index_id = i.index_id AND stat.[object_id] = i.[object_id] LEFT JOIN sys.partitions AS p WITH (NOLOCK) ON p.object_id = i.object_id AND p.index_id = i.index_id AND p.partition_number = 2 WHERE i.[type] IN (0, 1, 2)This is the query for This is the query including partition numbers, it's worth noting that if this is run in SSMS duplicates may occur but this is based on having multiple rows with different states, these will then be filtered out by the API as it only looks for SELECT CASE i.[type] WHEN 0 THEN CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], ' (HEAP)') ELSE CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], '.', i.[name]) END AS IndexFullName , stat.fragmentation, stat.page_count, 0 AS fragmentation_CI, 0 AS [state], i.[type] AS IndexType , CASE WHEN p.index_id IS NULL THEN '.' ELSE ' (Partition Number: ' + CAST(stat.partition_number AS varchar(5)) + ').' END AS partitionNumber FROM sys.indexes AS i WITH (NOLOCK) INNER JOIN sys.tables AS t WITH (NOLOCK) ON t.[object_id] = i.[object_id] AND t.is_ms_shipped = 0 INNER JOIN ( SELECT [object_id], index_id, SUM(ps.avg_fragmentation_in_percent) AS fragmentation, SUM(ps.page_count) AS page_count , ps.partition_number FROM sys.dm_db_index_physical_stats(DB_ID(), NULL, NULL, NULL, NULL) ps WHERE ps.index_level = 0 AND ps.alloc_unit_type_desc = 'IN_ROW_DATA' GROUP BY [object_id], index_id, ps.partition_number ) stat ON stat.index_id = i.index_id AND stat.[object_id] = i.[object_id] LEFT JOIN sys.partitions AS p WITH (NOLOCK) ON p.object_id = i.object_id AND p.index_id = i.index_id AND p.partition_number = 2 WHERE i.[type] IN (0, 1, 2, 7) UNION SELECT CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], '.', i.[name]) AS IndexFullName , 0, 0, SUM( (ISNULL(rg.deleted_rows, 1) * 100) /CASE WHEN rg.total_rows = 0 THEN 1 ELSE rg.total_rows END ), rg.[state], i.[type] , CASE WHEN p.index_id IS NULL THEN '.' ELSE ' (Partition Number: ' + CAST(rg.partition_number AS varchar(5)) + ').' END AS partitionNumber FROM sys.column_store_row_groups AS rg WITH (NOLOCK) INNER JOIN sys.indexes AS i WITH (NOLOCK) ON i.index_id = rg.index_id and i.[object_id] = rg.[object_id] INNER JOIN sys.tables AS t WITH (NOLOCK) ON t.[object_id] = rg.[object_id] AND t.is_ms_shipped = 0 LEFT JOIN sys.partitions AS p WITH (NOLOCK) ON p.object_id = i.object_id AND p.index_id = i.index_id GROUP BY rg.[object_id], i.[name], t.[name], t.[schema_id], rg.[state], i.[type], p.index_id, rg.partition_number ResultsYou can see in my test I get four results for that table, one for each partition that is fragmented above the threshold (0 in this example) These are the same SQL queries just formatted for readability: # older versions - no columnstore
SELECT CASE i.[type] WHEN 0 THEN CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], ' (HEAP)') ELSE CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], '.', i.[name]) END AS IndexFullName
, stat.fragmentation, stat.page_count, i.[type] AS IndexType --, stat.partition_number
, CASE WHEN p.index_id IS NULL THEN '.'
ELSE ' (Partition Number: ' + CAST(stat.partition_number AS varchar(5)) + ').'
END AS partitionNumber
FROM sys.indexes AS i WITH (NOLOCK)
INNER JOIN sys.tables AS t WITH (NOLOCK) ON t.[object_id] = i.[object_id]
AND t.is_ms_shipped = 0
INNER JOIN
(
SELECT [object_id], index_id, SUM(ps.avg_fragmentation_in_percent) AS fragmentation, SUM(ps.page_count) AS page_count
, ps.partition_number
FROM sys.dm_db_index_physical_stats(DB_ID(), NULL, NULL, NULL, NULL) ps
WHERE ps.index_level = 0
AND ps.alloc_unit_type_desc = 'IN_ROW_DATA'
GROUP BY [object_id], index_id , ps.partition_number
) stat
ON stat.index_id = i.index_id
AND stat.[object_id] = i.[object_id]
LEFT JOIN sys.partitions AS p WITH (NOLOCK)
ON p.object_id = i.object_id
AND p.index_id = i.index_id
AND p.partition_number = 2
WHERE i.[type] IN (0, 1, 2)
# 2012+ inc columnstore
SELECT
CASE i.[type]
WHEN 0 THEN CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], ' (HEAP)')
ELSE CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], '.', i.[name])
END AS IndexFullName
, stat.fragmentation, stat.page_count, 0 AS fragmentation_CI, 0 AS [state], i.[type] AS IndexType
, CASE
WHEN p.index_id IS NULL THEN '.'
ELSE ' (Partition Number: ' + CAST(stat.partition_number AS varchar(5)) + ').'
END AS partitionNumber
FROM sys.indexes AS i WITH (NOLOCK)
INNER JOIN sys.tables AS t WITH (NOLOCK)
ON t.[object_id] = i.[object_id]
AND t.is_ms_shipped = 0
INNER JOIN
(
SELECT [object_id], index_id, SUM(ps.avg_fragmentation_in_percent) AS fragmentation, SUM(ps.page_count) AS page_count
, ps.partition_number
FROM sys.dm_db_index_physical_stats(DB_ID(), NULL, NULL, NULL, NULL) ps
WHERE ps.index_level = 0 AND ps.alloc_unit_type_desc = 'IN_ROW_DATA'
GROUP BY [object_id], index_id, ps.partition_number
) stat
ON stat.index_id = i.index_id
AND stat.[object_id] = i.[object_id]
LEFT JOIN sys.partitions AS p WITH (NOLOCK)
ON p.object_id = i.object_id
AND p.index_id = i.index_id
AND p.partition_number = 2
WHERE i.[type] IN (0, 1, 2, 7)
UNION
SELECT CONCAT(SCHEMA_NAME(t.schema_id), '.', t.[name], '.', i.[name]) AS IndexFullName
, 0, 0, SUM( (ISNULL(rg.deleted_rows, 1) * 100) /CASE WHEN rg.total_rows = 0 THEN 1 ELSE rg.total_rows END ), rg.[state], i.[type]
, CASE
WHEN p.index_id IS NULL THEN '.'
ELSE ' (Partition Number: ' + CAST(rg.partition_number AS varchar(5)) + ').'
END AS partitionNumber
FROM sys.column_store_row_groups AS rg WITH (NOLOCK)
INNER JOIN sys.indexes AS i WITH (NOLOCK)
ON i.index_id = rg.index_id
and i.[object_id] = rg.[object_id]
INNER JOIN sys.tables AS t WITH (NOLOCK)
ON t.[object_id] = rg.[object_id]
AND t.is_ms_shipped = 0
LEFT JOIN sys.partitions AS p WITH (NOLOCK)
ON p.object_id = i.object_id
AND p.index_id = i.index_id
GROUP BY rg.[object_id], i.[name], t.[name], t.[schema_id], rg.[state], i.[type], p.index_id, rg.partition_number |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
If I run the following to check my local pubs database for fragmentation (
config.jsonlowers the threshold to 0) I get three rows per index as it is a partitioned table.If you add
sp.partition_numberinto the query that runs this check you'll be able to see which partitions are fragmented and then this could be appended to the message returnedThis was originally posted on the
SQLServerPSModulerepo but suggested I post heremicrosoft/SQLServerPSModule#68
The text was updated successfully, but these errors were encountered: