postgresql - Dedicated database server heavy iowait spikes -


We have a dedicated database server that runs PostgreSQL 8.3 on Linux Debian. The database is being routinely inquiries for a lot of data, while updates / insertions are often done. Databases are not a response to a short period (like 10 seconds) and then it can be used for normal performance Goes in again flows.

What I saw through the top is that the time that remains for as it is during a iowait nail that is, until the database does not respond, the pdflush becomes active at the same time. So my idea is that the PDF has to write back data from the cache memory space on the disk based on the messy page and background ratio. The rest of the time, when PostGraceCalclass works normally, no iPhite is running because pdflush is not active. The values ​​for my VM are as follows: 

  dirty_background_ratio = 5 dirty_ratio = 10 dirty_expire_centisecs = 3000

My meminfo: 

 < Code: MemTotal: 12403212 kB MemFree: 1779684 kB Buffer: 253,284 kB Cached: 9,076,132 kB SwapCached: 0 kB Active: 7,298,316 kB Not in use: 2,555,240 kB SwapTotal: 7,815,544 kB SwapFree: 7,814,884 kB Nasty: 1804 kB writeback: 0 kB AnonPages: 495,028 kB Mapped: 3142164 KB Slab: 280588 KB Acceptable: 265284 KB Acceptance: 15304 KB Pagetable: 422 9 80 kb NFS- Installable: 0 KB bounce: 0 KB typebackmp: 0 KB commission limit: 14017148 KB CATED_AS: 3890832 KB VALOCOCAL: 34359738367 KB VALUCUES: 304188 KB VMLLCANK: 34359433983 kB HugePages_Total: 0 HugePages_Free: 0 HugePages_Rsvd: 0 HugePages_Surp: 0 Hugepagesize: 2048 kB   
 I'm thinking of making changes The uration, which remains in a dirty page memory (dirty_expire_centisecs), as iowait nails to split evenly in time (more regular pdflush phones so as to write less amount of data on disk ). Any other proposed solution?   
 
 
 IO spike is likely to occur when postcrescale checkpointing is done. You can verify that from you and see if they match the lack of server response.  
 If such a thing is tuning,  checkpoints_segments  and  checkpoint_completion_target  are likely to help see more about this and about the doctor.

Comments

Post a Comment

Popular posts from this blog

Python SQLAlchemy：AttributeError: Neither 'Column' object nor 'Comparator' object has an attribute 'schema' -

I tried to create a new database in my project, but when I run the script, I get this error, I have another project, this type of definition worked before, but now it gets a single error. I am using Python 2.7.8 and the version of SQLAlchemy module is 0.9.8. By the way, using a project flask- SQLAlchemy, it works well, I'm confused. Traceback information is as follows: Traceback (most recent call final): File "D: / Projects / IOM / db_create.py", line 4, & lt; Module & gt; From the models the base file "D: \ Projects \ OO-IM \ models.py", in line 15, in the & lt; Module & gt; Column ( 'followed_id', int (), ForeignKey ( 'user.id')) file "C: \ Python27 \ lib \ site-packages \ SQLAlchemy \ SQL \ schema.py", line 369, __new__ schema = metadata. Schema file "C: \ Python27 \ lib \ site-packages \ SQLAlchemy \ SQL \ elements.py", line 662, AttributeError in __getattr__ key): Neither 'column' object nor ...
Read more

java - How not to audit a join table and related entities using Hibernate Envers? -

I use Eners to be hibernate to audit my institutions. I have an audit entity, Foo , which includes a list & lt; Bars & gt; However, I do not want to audit the bar institutions, in the form of properties, I have written that: @inti @edicated public class Foo {@JoinTable (name = "T_FOO_BAR", joinColumns = @JoinColumn (name = "FOO_ID"), inverseJoinColumns = @JoinColumn (Name = "BAR_ID")) @ManyToMany (Cascade = Persist) @ Indited (targetAuditMode = RelationTargetAuditMode.NOT_AUDITED ) Public listing & lt; Bars & gt; GetBars () {Return bars; }} Now, I want to retrieve an amendment of foo : AuditReader reader = AuditReaderFactory.get (GetEntityManager ()); FU modification = (FU) reader.CreativeTeA () .entensEtrevision (Foo Class, 42) .getSingleResult (); Unfortunately, when I want to retrieve all the data (i.e. when it loads lazy times ), error me ORA-00942: table or view is not present , because Tried to query this: Sel...
Read more

mongodb - CakePHP paginator ignoring order, but only for certain values -

While pagging my user model in kppp, I can sort with some values, but not others. For example, I can order results from created or email , for example, but username or reputation seems to come back order by arbitrary command for example with a list of users: $ this-> paginate = array ('condition' => array ( 'User.is_active' = & gt; true), 'border' => 24, 'order' => array ('User.created' => DESC)); Works as expected, but $ this - & gt; Paged = array ('conditions' = & gt; array (' user.isIactive '= & gt; true),' border '= & gt; 24,' command '= & gt; array (' user repeats' = ' & Gt; DESC ')); No, no. I first thought that this could be a database issue, but when I search the database directly, sorted as expected. Note: I am using Mangodebi with the Mongo DB plugin of Ichikawa for the KPPHP; Pagnetizing users are used to work properl...
Read more