Sender | Message | Time |
---|---|---|
23 Apr 2024 | ||
Tim P | * I've been talking (hopefully not erroneously) using phrases like "teed up in memory", and "tasks the scheduler is thinking about".
If it's a typing/taxi/subprocess pool as opposed to a filing system (the rest of the graph) this works well - if it's a pool of water then it's not unreasonable to expect it to be a contiguous thing, and it's not: Extending that analogy leads to task puddles. For Pythony people this is probably not a problem, but I don't know if other languages use the word pool in the same way.
How about loaded/archived ? I think this hints at the relationship between the pool and the graph... You get a 🐸 for mention spawning to a pool: Presumably 🐸 + 🔥 (flame wars) = 🦎 (salamander) | 08:19:44 |
oliver sanders | There's a cylc-flow issue which touches on this topic. | 08:23:53 |
oliver sanders | I'm not sure we actually need to expose the task pool and spawn terminologies to users. For the n-window, we only need to explain what n=0 means and how the n=<x> model extends that. For stalls, it's easy enough to say that Cylc could not proceed further through the graph (no need for pool terminology). The difference in task matching is something that we plan to address in the near future. | 08:30:44 |
Ronnie Dutta | Codecov failures: thoughts on adding the Codecov upload token in plaintext to the workflow to ensure it works? The Codecov maintainers have suggested you can do this, the worst that could happen is a malicious actor will upload false coverage 😱 | 10:09:00 |
Ronnie Dutta | * Codecov failures: thoughts on adding the Codecov upload token in plaintext to the workflow to ensure it works? The Codecov maintainers have suggested you can do this, the worst that could happen is a malicious actor could upload false coverage 😱 | 10:09:22 |
Hilary Oliver | Yeah, I'd forgotten. It's this one: https://github.com/cylc/cylc-flow/issues/3647 That issue has become pretty unwieldy. At some point we'll need to go through it, see what's been done already, and maybe supersede it with some smaller self-contained issues. | 23:15:08 |
24 Apr 2024 | ||
Hilary Oliver |
I more or less agree on "spawn" - it's not too hard to avoid that, but we can't avoid calling the task pool something.
Exactly, that's what I'm talking about! The term "n-window" doesn't really mean anything to the scheduler, it's a GUI thing. But nevertheless, the GUI n-window is centered on the scheduler task pool. So what do we call that? "Active window" is problematic for the reasons above. I think "scheduler task pool" is fine (for the other reasons above) but I'm open to other ideas if you have any. We have to call it something! We really need to clean up the documentation in this respect before doing the big 8.3.0 evangelising of the new intervention methods etc. | 02:07:25 |
Hilary Oliver | *
I more or less agree on "spawn" - it's not too hard to avoid that, but we can't avoid calling the task pool something.
Exactly, that's what I'm talking about! The term "n-window" doesn't really mean anything to the scheduler, it's a GUI thing. But nevertheless, the GUI n-window is centered on the scheduler task pool. So what do we call that? What exactly is the "pool" of tasks displayed by "Active window" is problematic for the reasons above. I think "scheduler task pool" is fine (for the other reasons above) but I'm open to other ideas if you have any. But we have to call it something! We really need to clean up the documentation in this respect before going all out on evangelizing 8.3.0 intervention methods etc. | 02:09:06 |
Hilary Oliver |
What I was getting at here is that stalls are caused by tasks that are retained in the the scheduler task pool, which is very much in the face of users as the basis of the n-window. And so failed tasks (e.g.) will always be visible regardless of the value of
Unfortunately, that ain't gonna be done in time for 8.3.0. I need to be doing migration seminars at NIWA (and possibly elsewhere) as soon as we release 8.3.0. And anyway, even if we can make all commands task pool agnostic, that won't eliminate the central reason for the user-facing nature of the task pool: it is "all the tasks that the scheduler is currently actively managing", and it is the basis/center of the GUI n-window. | 02:19:00 |
Hilary Oliver | *
What I was getting at here is that stalls are caused by tasks that are retained in the the scheduler task pool, which is very much in the face of users as the basis of the n-window. The fact that they are retained in task pool is what keeps them visible regardless of how far back in the graph they are.
Unfortunately that ain't gonna be done in time for 8.3.0. I need to be doing migration seminars at NIWA (and possibly elsewhere) as soon as we release 8.3.0. But in any case, even if we can make all commands entirely task pool agnostic, that won't eliminate the central reason for the user-facing nature of the task pool: it is the basis/center of the GUI n-window. | 02:24:25 |
Hilary Oliver | proposal - for 8.3.0 docsCan we agree on this: (A) We define the scheduler task pool as:
(B) We ditch the word "spawn":
Feel free to suggest other names for "task pool", but we have to call it something in order to describe what Cylc does (to users!). | 02:29:57 |
Hilary Oliver | * proposal - for 8.3.0 docsCan we agree on this: (A) We define the scheduler task pool as:
(B) We ditch the word "spawn":
Feel free to suggest other names for "task pool", but we have to call it something in order to describe what Cylc does (to users!). | 02:31:38 |
Hilary Oliver | * proposal - for 8.3.0 docsCan we agree on this: (A) We define the scheduler task pool as:
(B) We ditch the word "spawn":
Feel free to suggest other names for "task pool", but we have to call it something in order to describe what Cylc does (to users!). I'm doubtful there's anything better that we haven't thought of yet. | 02:32:24 |
Hilary Oliver | *
I more or less agree on "spawn" - it's not too hard to avoid that, but we can't avoid calling the task pool something.
Exactly, that's what I'm talking about! The term "n-window" doesn't really mean anything to the scheduler, it's a GUI thing. But nevertheless, the GUI n-window is centered on the scheduler task pool. So what do we call that? What exactly is the "pool" of tasks displayed by "Active window" is problematic for the reasons above. I think "scheduler task pool" is fine (for the other reasons above) but I'm open to other ideas if you have any. But we have to call it something! We really need to clean up the documentation in this respect before going all out on evangelizing 8.3.0 intervention methods etc. | 02:33:37 |
Hilary Oliver | *
What I was getting at here is that stalls are caused by tasks that are retained in the the scheduler task pool, which is very much in the face of users as the basis of the n-window. The fact that they are retained in task pool is what keeps them visible regardless of how far back in the graph they are.
Unfortunately that ain't gonna be done in time for 8.3.0. I need to be doing migration seminars at NIWA (and possibly elsewhere) as soon as we release 8.3.0. And we need to have the documentation sorted out accordingly by them. But in any case, even if we can make all commands entirely task pool agnostic, that won't eliminate the central reason for the user-facing nature of the task pool: it is the basis/center of the GUI n-window. | 02:35:02 |
Hilary Oliver | * proposal - for 8.3.0 docsCan we agree on this: (A) We define the scheduler task pool as:
(B) We ditch the word "spawn":
Feel free to suggest other names for "task pool", but we have to call it something in order to describe what Cylc does (to users!). I'm doubtful there's anything better that we haven't thought of yet. | 02:37:45 |
Hilary Oliver | * proposal - for 8.3.0 docsCan we agree on this: (A) We define the scheduler task pool as:
(B) We ditch the word "spawn":
Feel free to suggest other names for "task pool", but we have to call it something in order to describe (to users) what Cylc does and what they see in the UIs. I'm doubtful there's anything better that we haven't thought of yet. | 02:38:35 |
Hilary Oliver | * proposal - for 8.3.0 docsCan we agree on this: (A) We define the scheduler task pool(*) as:
(B) We ditch the word "spawn":
(*) Feel free to suggest other names for "task pool", but we have to call it something in order to describe (to users) what Cylc does and what they see in the UIs. I'm doubtful there's anything better that we haven't thought of yet. | 02:42:55 |
Hilary Oliver | *
What I was getting at here is that stalls are caused by tasks that are retained in the the scheduler task pool, which is very much in the face of users as the basis of the n-window. The fact that they are retained in task pool is what keeps them visible regardless of how far back in the graph they are.
Unfortunately that ain't gonna be done in time for 8.3.0. I need to be doing migration seminars at NIWA (and possibly elsewhere) as soon as we release 8.3.0. And we need to have the documentation sorted out accordingly by then. But in any case, even if we can make all commands entirely task pool agnostic, that won't eliminate the central reason for the user-facing nature of the task pool: it is the basis/center of the GUI n-window. | 03:35:37 |
Hilary Oliver | I'm probably belabouring the point now, but IMO we can't really just call the task pool "n=0". That just begs the question, zero graph-edges out from what?. | 03:39:41 |
Hilary Oliver | * Final end-of-the day comment on this:
oliver sanders are you suggesting we call the task pool "n=0" instead of "the task pool"? I'm not entirely averse to that, although it seems a little nonsensical when the GUI is not involved ("n edges out from what?"). But even that would just be a different name for the same concept, which is, like or not, "the pool/group/whatever of tasks that the scheduler is actively managing at this moment" - that is what I don't think we can realistically avoid referring to (by whatever name) in the docs, for the reasons above. | 06:08:59 |
Hilary Oliver | * Final end-of-the day comment on this:
oliver sanders are you suggesting we call the task pool "n=0" instead of "the task pool"? I'm not entirely averse to that, although it seems a little nonsensical when the GUI is not involved ("n edges out from what?"). But even that would just bd a different name for the same concept, which is, "the pool/group/whatever of tasks that the scheduler is actively managing at this moment" - that is what I don't think we can realistically avoid referring to (by whatever name) in the docs, for the reasons above. | 06:09:41 |
Hilary Oliver | * Final end-of-the day comment on this:
oliver sanders are you suggesting we call the task pool "n=0" instead of "the task pool"? I'm not entirely averse to that, although it seems a little nonsensical when the GUI is not involved ("n edges out from what?"). But even that would just bd a different name for the same concept - i.e., "the pool/group/whatever of tasks that the scheduler is actively managing at this moment"_ . | 06:10:41 |
Ronnie Dutta | Heads up, there seems to be issues with cylc hub for jupyterhub versions in the range 4.1, related to XRSF cookie handling which means the UI may fail to load or just the introspection query (and therefore the mutations menu) | 16:25:15 |
oliver sanders | https://github.com/jupyterlab/jupyterlab/issues/16040 | 16:50:19 |
oliver sanders | https://github.com/jupyterhub/nbgitpuller/issues/344 | 16:51:12 |
Ronnie Dutta | Redacted or Malformed Event | 17:42:36 |
25 Apr 2024 | ||
Hilary Oliver | In reply to @revilo666:matrix.org Hey all, I want to get on with documentation tweaks for 8.3.0. I'll take the lack of response to this to mean no one had major objections or counterarguments. One adjustment though, on looking closer, "spawn" is the right user-facing term to describe how the flow advances when outputs are completed - see https://github.com/cylc/cylc-flow/pull/6067#discussion_r1580076467 As an aside, oliver sanders famously claims to dislike terminology debates (despite being the prime offender other than myself, - "it takes two to tango" as the saying goes!) but really it's unavoidable and fundamentally a good thing - how else can we settle on the best way to describe and explain these sometimes quite subtle concepts. | 20:52:19 |
26 Apr 2024 | ||
oliver sanders | I'm not getting the pressing urgency for a terminology rejig so haven't looked through this yet. Do you consider this a 8.3.0 blocker? We're getting together for a meeting on the 8th, would that be a good time to discuss this? | 08:05:08 |
Ronnie Dutta | https://github.com/jupyterhub/jupyterhub/issues/4800 | 13:01:10 |