This is a spin-off of #251792: Implement a locking framework for long operations for D7 only.

The issue is still the same: when a cache entry expire and need to be regenerated (think locale cache, menu cache, etc.), several regeneration can occur in parallel because we currently have no locking. This hurts badly websites that sustain high request rates, especially in cluster environments.

The D7 implementation will be very different from the D6 one:

  • The system will comply to our new pluggable paradigms.
  • We will strongly link caching and locking. This make sense because we are generally locking to prepare a cached object: if using an APC cache for example, that is local to each web node, we don't need to lock the whole cluster.
  • We will have a new (probably fluent) API.

Comments

kbahey’s picture

kbahey CreditAttribution: kbahey commented

Thanks Damien. Subscribing.

moshe weitzman’s picture

moshe weitzman CreditAttribution: moshe weitzman commented

Subscribe as well

birdmanx35’s picture

birdmanx35 CreditAttribution: birdmanx35 commented
Status: Active » Postponed

Perhaps we should wait for #363787: Plugins: Swappable subsystems for core?

moshe weitzman’s picture

moshe weitzman CreditAttribution: moshe weitzman commented
Status: Postponed » Active

We should not postpone in favor of extremely early stage patches that are only slightly related. We have substantial issues here that need review beyond just the mechanics of how we make something pluggable. The only answer to that question in terms of this issue is"we make pluggable using the standard practice in core" which is not handlers right now.

Also, please do not change status with a msg like "perhaps".

Jeremy’s picture

Jeremy CreditAttribution: Jeremy commented

Subscribing.

michtoen’s picture

michtoen CreditAttribution: michtoen commented

+1

slantview’s picture

slantview CreditAttribution: slantview commented

This issue seems to be idling a bit. what do we need to push this forward, and is it too late for D7 already?

CorniI’s picture

CorniI CreditAttribution: CorniI commented

subscribe, I'd like to see and review how you implement the locking mechanism :)

coltrane’s picture

coltrane
he/him
CreditAttribution: coltrane commented

#251792: Implement a locking framework for long operations has a patch for D7 as RTBC, is this issue here still needed?

Damien Tournoud’s picture

Damien Tournoud CreditAttribution: Damien Tournoud commented
Status: Active » Closed (duplicate)

Yep, let's close that.