Suddenly slow RAID performance

Question

We recently noticed our database queries have been taking much longer than usual to run. After some investigation, it looks like we're getting very slow disk reads.

We've run into a similar problem in the past caused by the RAID controller initiating a relearn cycle on the BBU and switching over to write-through. It doesn't look like this is the case this time.

I ran bonnie++ a few times over the course of a couple days. Here's the results:

The 22-82 M/s reads seem pretty abysmal. Running dd against the raw device for a few minutes shows anywhere from 15.8 MB/s to 225 MB/s reads (see update below). iotop doesn't indicate any other processes competing for IO so I'm not sure why the read speed is so variable.

The raid card is a MegaRAID SAS 9280 with 12 SAS drives (15k, 300GB) in RAID10 with an XFS filesystem (OS on two SSDs configured in RAID1). I don't see any S.M.A.R.T. alerts and the array doesn't appear to be degraded.

I've also run xfs_check and there don't appear to be any XFS consistency issues.

What should be the next investigative steps here?

Server specs

Ubuntu 12.04.5 LTS
128GB RAM
Intel(R) Xeon(R) CPU E5-2643 0 @ 3.30GHz

Output of xfs_repair -n:

Phase 1 - find and verify superblock...
Phase 2 - using internal log
        - scan filesystem freespace and inode maps...
        - found root inode chunk
Phase 3 - for each AG...
        - scan (but don't clear) agi unlinked lists...
        - process known inodes and perform inode discovery...
        - agno = 0
        - agno = 1
        - agno = 2
        - agno = 3
        - process newly discovered inodes...
Phase 4 - check for duplicate blocks...
        - setting up duplicate extent list...
        - check for inodes claiming duplicate blocks...
        - agno = 1
        - agno = 3
        - agno = 2
        - agno = 0
No modify flag set, skipping phase 5
Phase 6 - check inode connectivity...
        - traversing filesystem ...
        - traversal finished ...
        - moving disconnected inodes to lost+found ...
Phase 7 - verify link counts...
No modify flag set, skipping filesystem flush and exiting.

Output of megacli -AdpAllInfo -aAll:

                    Versions
                ================
Product Name    : LSI MegaRAID SAS 9280-4i4e
Serial No       : SV24919344
FW Package Build: 12.12.0-0124

                    Mfg. Data
                ================
Mfg. Date       : 12/06/12
Rework Date     : 00/00/00
Revision No     : 04B
Battery FRU     : N/A

                Image Versions in Flash:
                ================
FW Version         : 2.130.363-1846
BIOS Version       : 3.25.00_4.12.05.00_0x05180000
Preboot CLI Version: 04.04-020:#%00009
WebBIOS Version    : 6.0-51-e_47-Rel
NVDATA Version     : 2.09.03-0039
Boot Block Version : 2.02.00.00-0000
BOOT Version       : 09.250.01.219

                Pending Images in Flash
                ================
None

                PCI Info
                ================
Controller Id   : 0000
Vendor Id       : 1000
Device Id       : 0079
SubVendorId     : 1000
SubDeviceId     : 9282

Host Interface  : PCIE

ChipRevision    : B4

Link Speed       : 0
Number of Frontend Port: 0
Device Interface  : PCIE

Number of Backend Port: 8
Port  :  Address
0        5003048001c1e47f
1        0000000000000000
2        0000000000000000
3        0000000000000000
4        0000000000000000
5        0000000000000000
6        0000000000000000
7        0000000000000000

                HW Configuration
                ================
SAS Address      : 500605b005a6cbc0
BBU              : Present
Alarm            : Present
NVRAM            : Present
Serial Debugger  : Present
Memory           : Present
Flash            : Present
Memory Size      : 512MB
TPM              : Absent
On board Expander: Absent
Upgrade Key      : Absent
Temperature sensor for ROC    : Absent
Temperature sensor for controller    : Absent


                Settings
                ================
Current Time                     : 14:58:51 7/11, 2016
Predictive Fail Poll Interval    : 300sec
Interrupt Throttle Active Count  : 16
Interrupt Throttle Completion    : 50us
Rebuild Rate                     : 30%
PR Rate                          : 30%
BGI Rate                         : 30%
Check Consistency Rate           : 30%
Reconstruction Rate              : 30%
Cache Flush Interval             : 4s
Max Drives to Spinup at One Time : 4
Delay Among Spinup Groups        : 2s
Physical Drive Coercion Mode     : Disabled
Cluster Mode                     : Disabled
Alarm                            : Enabled
Auto Rebuild                     : Enabled
Battery Warning                  : Enabled
Ecc Bucket Size                  : 15
Ecc Bucket Leak Rate             : 1440 Minutes
Restore HotSpare on Insertion    : Disabled
Expose Enclosure Devices         : Enabled
Maintain PD Fail History         : Enabled
Host Request Reordering          : Enabled
Auto Detect BackPlane Enabled    : SGPIO/i2c SEP
Load Balance Mode                : Auto
Use FDE Only                     : No
Security Key Assigned            : No
Security Key Failed              : No
Security Key Not Backedup        : No
Default LD PowerSave Policy      : Controller Defined
Maximum number of direct attached drives to spin up in 1 min : 120
Auto Enhanced Import             : No
Any Offline VD Cache Preserved   : No
Allow Boot with Preserved Cache  : No
Disable Online Controller Reset  : No
PFK in NVRAM                     : No
Use disk activity for locate     : No
POST delay           : 90 seconds
BIOS Error Handling              : Stop On Errors
Current Boot Mode         :Normal
                Capabilities
                ================
RAID Level Supported             : RAID0, RAID1, RAID5, RAID6, RAID00, RAID10, RAID50, RAID60, PRL 11, PRL 11 with spanning, SRL 3 supported, PRL11-RLQ0 DDF layout with no span, PRL11-RLQ0 DDF layout with span
Supported Drives                 : SAS, SATA

Allowed Mixing:

Mix in Enclosure Allowed
Mix of SAS/SATA of HDD type in VD Allowed

                Status
                ================
ECC Bucket Count                 : 0

                Limitations
                ================
Max Arms Per VD          : 32
Max Spans Per VD         : 8
Max Arrays               : 128
Max Number of VDs        : 64
Max Parallel Commands    : 1008
Max SGE Count            : 80
Max Data Transfer Size   : 8192 sectors
Max Strips PerIO         : 42
Max LD per array         : 16
Min Strip Size           : 8 KB
Max Strip Size           : 1.0 MB
Max Configurable CacheCade Size: 0 GB
Current Size of CacheCade      : 0 GB
Current Size of FW Cache       : 350 MB

                Device Present
                ================
Virtual Drives    : 2
  Degraded        : 0
  Offline         : 0
Physical Devices  : 16
  Disks           : 14
  Critical Disks  : 0
  Failed Disks    : 0

                Supported Adapter Operations
                ================
Rebuild Rate                    : Yes
CC Rate                         : Yes
BGI Rate                        : Yes
Reconstruct Rate                : Yes
Patrol Read Rate                : Yes
Alarm Control                   : Yes
Cluster Support                 : No
BBU                             : Yes
Spanning                        : Yes
Dedicated Hot Spare             : Yes
Revertible Hot Spares           : Yes
Foreign Config Import           : Yes
Self Diagnostic                 : Yes
Allow Mixed Redundancy on Array : No
Global Hot Spares               : Yes
Deny SCSI Passthrough           : No
Deny SMP Passthrough            : No
Deny STP Passthrough            : No
Support Security                : No
Snapshot Enabled                : No
Support the OCE without adding drives : Yes
Support PFK                     : Yes
Support PI                      : No
Support Boot Time PFK Change    : No
Disable Online PFK Change       : No
PFK TrailTime Remaining         : 0 days 0 hours
Support Shield State            : No
Block SSD Write Disk Cache Change: No

                Supported VD Operations
                ================
Read Policy          : Yes
Write Policy         : Yes
IO Policy            : Yes
Access Policy        : Yes
Disk Cache Policy    : Yes
Reconstruction       : Yes
Deny Locate          : No
Deny CC              : No
Allow Ctrl Encryption: No
Enable LDBBM         : No
Support Breakmirror  : No
Power Savings        : No

                Supported PD Operations
                ================
Force Online                            : Yes
Force Offline                           : Yes
Force Rebuild                           : Yes
Deny Force Failed                       : No
Deny Force Good/Bad                     : No
Deny Missing Replace                    : No
Deny Clear                              : No
Deny Locate                             : No
Support Temperature                     : Yes
NCQ                                     : No
Disable Copyback                        : No
Enable JBOD                             : No
Enable Copyback on SMART                : No
Enable Copyback to SSD on SMART Error   : Yes
Enable SSD Patrol Read                  : No
PR Correct Unconfigured Areas           : Yes
Enable Spin Down of UnConfigured Drives : Yes
Disable Spin Down of hot spares         : No
Spin Down time                          : 30
T10 Power State                         : No
                Error Counters
                ================
Memory Correctable Errors   : 0
Memory Uncorrectable Errors : 0

                Cluster Information
                ================
Cluster Permitted     : No
Cluster Active        : No

                Default Settings
                ================
Phy Polarity                     : 0
Phy PolaritySplit                : 0
Background Rate                  : 30
Strip Size                       : 256kB
Flush Time                       : 4 seconds
Write Policy                     : WB
Read Policy                      : Adaptive
Cache When BBU Bad               : Disabled
Cached IO                        : No
SMART Mode                       : Mode 6
Alarm Disable                    : Yes
Coercion Mode                    : None
ZCR Config                       : Unknown
Dirty LED Shows Drive Activity   : No
BIOS Continue on Error           : 0
Spin Down Mode                   : None
Allowed Device Type              : SAS/SATA Mix
Allow Mix in Enclosure           : Yes
Allow HDD SAS/SATA Mix in VD     : Yes
Allow SSD SAS/SATA Mix in VD     : No
Allow HDD/SSD Mix in VD          : No
Allow SATA in Cluster            : No
Max Chained Enclosures           : 16
Disable Ctrl-R                   : Yes
Enable Web BIOS                  : Yes
Direct PD Mapping                : No
BIOS Enumerate VDs               : Yes
Restore Hot Spare on Insertion   : No
Expose Enclosure Devices         : Yes
Maintain PD Fail History         : Yes
Disable Puncturing               : No
Zero Based Enclosure Enumeration : No
PreBoot CLI Enabled              : Yes
LED Show Drive Activity          : Yes
Cluster Disable                  : Yes
SAS Disable                      : No
Auto Detect BackPlane Enable     : SGPIO/i2c SEP
Use FDE Only                     : No
Enable Led Header                : No
Delay during POST                : 0
EnableCrashDump                  : No
Disable Online Controller Reset  : No
EnableLDBBM                      : No
Un-Certified Hard Disk Drives    : Allow
Treat Single span R1E as R10     : No
Max LD per array                 : 16
Power Saving option              : Don't Auto spin down Configured Drives
Max power savings option is  not allowed for LDs. Only T10 power conditions are to be used.
Default spin down time in minutes: 30
Enable JBOD                      : No
TTY Log In Flash                 : No
Auto Enhanced Import             : No
BreakMirror RAID Support         : No
Disable Join Mirror              : No
Enable Shield State              : No
Time taken to detect CME         : 60s

Output of megacli -AdpBbuCmd -GetBbuSTatus -aAll:

BBU status for Adapter: 0

BatteryType: iBBU
Voltage: 4068 mV
Current: 0 mA
Temperature: 30 C
Battery State: Optimal
BBU Firmware Status:

  Charging Status              : Charging
  Voltage                                 : OK
  Temperature                             : OK
  Learn Cycle Requested                   : No
  Learn Cycle Active                      : No
  Learn Cycle Status                      : OK
  Learn Cycle Timeout                     : No
  I2c Errors Detected                     : No
  Battery Pack Missing                    : No
  Battery Replacement required            : No
  Remaining Capacity Low                  : No
  Periodic Learn Required                 : No
  Transparent Learn                       : No
  No space to cache offload               : No
  Pack is about to fail & should be replaced : No
  Cache Offload premium feature required  : No
  Module microcode update required        : No


GasGuageStatus:
  Fully Discharged        : No
  Fully Charged           : No
  Discharging             : Yes
  Initialized             : Yes
  Remaining Time Alarm    : No
  Discharge Terminated    : No
  Over Temperature        : No
  Charging Terminated     : No
  Over Charged            : No
  Relative State of Charge: 88 %
  Charger System State: 49169
  Charger System Ctrl: 0
  Charging current: 512 mA
  Absolute state of charge: 87 %
  Max Error: 4 %

Exit Code: 0x00

Output of megacli -LDInfo -Lall -aAll:

Adapter 0 -- Virtual Drive Information:
Virtual Drive: 0 (Target Id: 0)
Name                :
RAID Level          : Primary-1, Secondary-0, RAID Level Qualifier-0
Size                : 111.281 GB
Sector Size         : 512
Mirror Data         : 111.281 GB
State               : Optimal
Strip Size          : 256 KB
Number Of Drives    : 2
Span Depth          : 1
Default Cache Policy: WriteBack, ReadAhead, Direct, No Write Cache if Bad BBU
Current Cache Policy: WriteBack, ReadAhead, Direct, No Write Cache if Bad BBU
Default Access Policy: Read/Write
Current Access Policy: Read/Write
Disk Cache Policy   : Disk's Default
Encryption Type     : None
Is VD Cached: No


Virtual Drive: 1 (Target Id: 1)
Name                :
RAID Level          : Primary-1, Secondary-0, RAID Level Qualifier-0
Size                : 1.633 TB
Sector Size         : 512
Mirror Data         : 1.633 TB
State               : Optimal
Strip Size          : 256 KB
Number Of Drives per span:2
Span Depth          : 6
Default Cache Policy: WriteBack, ReadAhead, Direct, Write Cache OK if Bad BBU
Current Cache Policy: WriteBack, ReadAhead, Direct, Write Cache OK if Bad BBU
Default Access Policy: Read/Write
Current Access Policy: Read/Write
Disk Cache Policy   : Disk's Default
Encryption Type     : None
Is VD Cached: No

Update: Per Andrew's advice, I ran dd for a few minutes to see what kind of throughput I'd get on raw disk reads:

dd if=/dev/sdb of=/dev/null bs=256k
19701+0 records in
19700+0 records out
5164236800 bytes (5.2 GB) copied, 202.553 s, 25.5 MB/s

Results of other runs, with highly variable throughput:

18706857984 bytes (19 GB) copied, 1181.51 s, 15.8 MB/s
20923023360 bytes (21 GB) copied, 388.137 s, 53.9 MB/s
21205876736 bytes (21 GB) copied, 55.5997 s, 381 MB/s
25391005696 bytes (25 GB) copied, 153.903 s, 165 MB/s

Update 2: Output of megacli -PDlist -aall: https://gist.github.com/danpelota/3fca1e5f90a1f358c2d52a49bfb08ef0

Try reading directly from the disk device. Something like `dd if=/dev/sdb of=/dev/null bs=256k` and see what bandwidth you get. Just make sure you don't write to the disk device unless you want to restore your filesystem(s) from backup... — Andrew Henle, Jul 11 '16 at 17:33
Great idea @AndrewHenle . I updated the question with the results (25.5 MB/s). — danpelota, Jul 11 '16 at 17:43
Ouch. 25 MB/s is pretty poor for that setup. What bandwidth do you get off the SSD array? — Andrew Henle, Jul 11 '16 at 18:07
Yeah, it's pretty brutal. Getting 1.1 GB/s on the SSD. Should I be concerned that the Bbu is showing a status of "Charging"? — danpelota, Jul 11 '16 at 18:16
I'm not familiar enough with that RAID controller to know. Before your performance dropped, did the BBU show that? And I'd think, given your configuration, that the state of the BBU shouldn't have any impact on reads at all. Have you looked through your logs for any problems? — Andrew Henle, Jul 11 '16 at 18:58
@AndrewHenle - Good question - I'm honestly not sure what the BBU was showing before performance dropped. Don't see anything in `syslog` or the raid software's event log, and `iotop` isn't showing any processing using significant IO. — danpelota, Jul 12 '16 at 01:03
How does smart look like on that SASes (`smartctl -a -d sat+megaraid,21 /dev/sda`)? Show us also: `megacli -PDlist -aall ` — Michal Sokolowski, Jul 12 '16 at 04:42
@MichalSokolowski: Updated with link to the output of PDlist (exceeds character limit of post). Don't see any SMART alerts or indication of drives in prefailure. Running `smartctl -a -d sat+megaraid,10 /dev/sdb` gives me "A mandatory SMART command failed: exiting.". Running `smartctl -a -d megaraid,10 /dev/sdb` returns basic controller information (vendor, product, unit id), but exits with "Device does not support SMART". Perhaps I'm doing something wrong? — danpelota, Jul 12 '16 at 12:28
@Dan, I looked in details at `megacli -PDlist -aall` output and I do not see anything obviously wrong there, `smartctl -a -d sat+megaraid,10 /dev/sdb` was only an example, checking SMART counters is worth the shot IMHO, SMART alerts rarely worked for me, first find correct module settings for your drives using `smartctl --scan`, then replace part `sat+megaraid,21` leave `/dev/sda` untouched, it should not change anything anyway. I used that way on `ServeRAID M5015 SAS/SATA Controller`, it looks like same sh*t though. Here's an example: http://pastebin.com/WYb8Utxr — Michal Sokolowski, Jul 12 '16 at 15:00
@MichalSokolowski: Thanks for the guidance. `smartctl --scan` is returning the device type, but not much else (https://gist.github.com/danpelota/266ec3314a22606b99b5e8df06d2e8a9). Interestingly, I'm now getting wildly varying read speeds: Running `dd` at various times over the last day has gotten me anywhere between 20 MB/s and 225 MB/s (still slow for this array). Again, `iotop` doesn't show any other processes competing for IO... — danpelota, Jul 12 '16 at 18:55
@Dan, for me such behavior is still consisting with "prefailing" disk(s). It's possible that some disk almost always recovering your reads which slows down whole process. SMART stats are crucial to approve/reject my theory. Again, for me megaraid counters were worthless, it didn't changed until disk died completely (the state when it doesn't start at all), I've used similar SASes in terms of vendor and size. — Michal Sokolowski, Jul 13 '16 at 07:43
@MichalSokolowski: Makes sense. Got the correct device numbers and ran `smartctl` on the individual disks. Looks like the non-medium error count is huge for disk 18: https://gist.github.com/danpelota/83b54854aa5af2e351ed71af5c8ebbf5 — danpelota, Jul 13 '16 at 12:59
@MichalSokolowski: That did it! I noticed the non-medium error count on device 18, disabled that drive via `megacli`, and now I'm getting well over 3GB reads. Thanks so much. If you want to post your process as an answer I'll accept it. Otherwise I'll do a write up and answer myself. — danpelota, Jul 13 '16 at 15:02
Glad you nailed it. Please feel free to do that. I'll up vote it up anyway. Reputation increase is only side effect. Please remember to make question answered. You did most the of job. :) I've read your SMART stats, but in current form they're useless too. I wanted to see actual SMART counters. Weirdly your `smartctl` doesn't show it like mine. There's actual answer for sure. — Michal Sokolowski, Jul 13 '16 at 15:47
@MichalSokolowski or danpelota can you please mark this question as solved (aka posting an answer)? — Broco, Jan 11 '17 at 09:38
If your RAID system can tolerate it, you might want to pull one drive at a time and put it in another system and make sure the SMART numbers look good. If it is having to relearn, it might be because of having to spend too much time dealing with bad blocks. Of course, if that is the cace, then I would consider replacing that drive. — SDsolar, Feb 21 '17 at 00:49
@Michael: Please Post your comment as answer. Credit belongs to you. And an answer is easier to find/reference than scrolling through comments. — Nils, Mar 19 '17 at 21:54
Do not use bonnie for device benchmarks. Use fio with file on filesystem or directly on block device. bonnie mostly useful for filesystem metadata benchmarking, and it usually unable to fully utilize (i.e. to benchmark) modern block devices. — George Shuklin, Mar 23 '17 at 10:20
We already had issues with the xfs file system with problems such as File system not showing the correct size even after running the xfs_check and xfs_repair. Can you paste the dmesg output of your server here? Also let us know the output of the below commands. `ioping /dev/partition` and `hdparm -tT /dev/partition` — Jaison V John, May 25 '17 at 03:28

score 7 · Accepted Answer · edited Jun 11 '20 at 10:02

As Michal pointed out in his comment, the issue was a "prefailing" disk. There were no red flags in the diagnostics from the megaraid controller and smartctl's SMART Health Status: was OK, but running smartctl on each disk revealed a huge Non-medium error count (I wrote a quick bash script to loop through each disk ID). Here's the relevant bits from the full output:

# Ran this for each individual disk on the /dev/sdb array:
smartctl -a -d megaraid,18  /dev/sdb

Error counter log:
           Errors Corrected by           Total   Correction     Gigabytes    Total
               ECC          rereads/    errors   algorithm      processed    uncorrected
           fast | delayed   rewrites  corrected  invocations   [10^9 bytes]  errors
read:    7950078        0         0   7950078    7950078        660.801           0
write:         0        0         0         0          0        363.247           0
verify:       12        0         0        12         12          0.002           0

Non-medium error count:  3253718

Every other drive showed a non-medium error count of 0, except for this one (disk ID 18). I identified the disk, swapped it with a new one, and am back to getting 3gbps reads.

According to smartmontools wiki:

The displayed error logs (if available) are displayed on separate lines:

write error counters

read error counters

verify error counters (only displayed if non-zero)

non-medium error counter (only a single number displayed). This represents the number of recoverable events other than write, read or verify errors.

error events are held in the "Last n error events" log page. The number of error event records held (i.e. "n") is vendor specific (e.g. up to 23 records are held for Hitachi 10K300 model disks). The contents of each error event record is in ASCII and vendor specific. The parameter code associated with each error event record indicates the relative time at which the error event occurred. A higher parameter code indicates that the error event occurred later in time. If this log page is not supported by the device then "Error Events logging not supported" is output. If this log page is supported and there are error event records then each one is prefixed by "Error event :" where is the parameter code.

I've hit it again, so I am making notes in your answer if you do not mind. — Michal Sokolowski, Dec 13 '17 at 07:59

score 0 · Answer 2 · edited Mar 28 '17 at 15:05

0

You need to verify the fragmentation of your drive :

xfs_db -r /dev/sdbx
frag

You'll have an answer like this :

actual 347954, ideal 15723, fragmentation factor 95.48%

If your fragmentation factor is high, you'll need to defrag your disk. (yes I know, like on Windows...) :/

To defrag your disk : xfs_fsr -v /dev/sdbx

edited Mar 28 '17 at 15:05

chicks

3,639
10
26
36

answered Mar 28 '17 at 13:27

Christophe Casalegno

41
4

score 0 · Answer 3 · answered Apr 20 '17 at 22:18

With the LSI's there are a few of things that are really important.

1) Flash the RAID firmware. You're a few revs off current.

2) Flash the firmware on the drives and make sure it's update to as well.

3) Update your driver. Based on the release notes on LSI's website, they just released a new driver at the end January.

Then you can re-run your tests to see if there is any change.

Suddenly slow RAID performance

Server specs

3 Answers3