Page MenuHomePhabricator

DB hardware for Reading Lists
Closed, DeclinedPublic

Description

Per the discussion in T164990, the x1 database cluster is close enough to full that a new service does not comfortably fit there. Reading Lists can temporarily live there with some restrictions but should eventually go to the to-be-established new cluster x2. That means new hardware needs to be procured - per T164990#3264314, probably 4 DB servers. This should be coordinated with the other would-be tenants of the new cluster.

(Not adding SRE hardware-requests yet as this is just a tracking task and not actionable right now; will probably happen in next annual budget planning.)

Event Timeline

Jhernandez subscribed.

I'll be checking in to see how we can make this actionable.

new cluster x2

When budgeting this year's resources, SRE decision (not only mine, Mark signed it off) was to budget to increase x1 servers rather than setting up a separate, dedicated cluster. That is coming for codfw on Q1 and for eqiad in Q3.

There was already a small increase of resources in late Q4.

@Tgr Can you specify the needs of the service to @jcrespo to see if they can be met with the planned increase?

@jcrespo where can we see the planned increase in detail? Thanks for chiming in!

I don't really have anything beyond what was projected in T164990. (The current load is around 1 read/sec an 0.1 write/sec, but the feature is not used on the web which would be the overwhelming majority of the load. OTOH there are some temporary limitations in place: no more than 10 lists or 100 list entries read in a single request - those would presumably be lifted if the DB permits.)

Jhernandez lowered the priority of this task from High to Medium.Jul 10 2018, 4:55 PM

Lowering to normal since it is not clear when the web will implement this and apps traffic don't seem to pose a problem right now.

As this hasn't come up in the past 2+ years.