Files
neon/libs/pageserver_api/src/controller_api.rs
Vlad Lazar 2cf47b1477 storcon: do az aware scheduling (#9083)
## Problem

Storage controller didn't previously consider AZ locality between
compute and pageservers
when scheduling nodes. Control plane has this feature, and, since we are
migrating tenants
away from it, we need feature parity to avoid perf degradations.

## Summary of changes

The change itself is fairly simple:
1. Thread az info into the scheduler
2. Add an extra member to the scheduling scores

Step (2) deserves some more discussion. Let's break it down by the shard
type being scheduled:

**Attached Shards**

We wish for attached shards of a tenant to end up in the preferred AZ of
the tenant since that
is where the compute is like to be. 

The AZ member for `NodeAttachmentSchedulingScore` has been placed
below the affinity score (so it's got the second biggest weight for
picking the node). The rationale for going
below the affinity score is to avoid having all shards of a single
tenant placed on the same node in 2 node
regions, since that would mean that one tenant can drive the general
workload of an entire pageserver.
I'm not 100% sure this is the right decision, so open to discussing
hoisting the AZ up to first place.

 **Secondary Shards**

We wish for secondary shards of a tenant to be scheduled in a different
AZ from the preferred one
for HA purposes.

The AZ member for `NodeSecondarySchedulingScore` has been placed first,
so nodes in different AZs
from the preferred one will always be considered first. On small
clusters, this can mean that all the secondaries
of a tenant are scheduled to the same pageserver, but secondaries don't
use up as many resources as the
attached location, so IMO the argument made for attached shards doesn't
hold.

Related: https://github.com/neondatabase/neon/issues/8848
2024-09-25 14:31:04 +01:00

399 lines
12 KiB
Rust

use std::collections::{HashMap, HashSet};
use std::fmt::Display;
use std::str::FromStr;
use std::time::{Duration, Instant};
/// Request/response types for the storage controller
/// API (`/control/v1` prefix). Implemented by the server
/// in [`storage_controller::http`]
use serde::{Deserialize, Serialize};
use utils::id::{NodeId, TenantId};
use crate::models::PageserverUtilization;
use crate::{
models::{ShardParameters, TenantConfig},
shard::{ShardStripeSize, TenantShardId},
};
#[derive(Serialize, Deserialize, Debug)]
#[serde(deny_unknown_fields)]
pub struct TenantCreateRequest {
pub new_tenant_id: TenantShardId,
#[serde(default)]
#[serde(skip_serializing_if = "Option::is_none")]
pub generation: Option<u32>,
// If omitted, create a single shard with TenantShardId::unsharded()
#[serde(default)]
#[serde(skip_serializing_if = "ShardParameters::is_unsharded")]
pub shard_parameters: ShardParameters,
#[serde(default)]
#[serde(skip_serializing_if = "Option::is_none")]
pub placement_policy: Option<PlacementPolicy>,
#[serde(flatten)]
pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
}
#[derive(Serialize, Deserialize)]
pub struct TenantCreateResponseShard {
pub shard_id: TenantShardId,
pub node_id: NodeId,
pub generation: u32,
}
#[derive(Serialize, Deserialize)]
pub struct TenantCreateResponse {
pub shards: Vec<TenantCreateResponseShard>,
}
#[derive(Serialize, Deserialize)]
pub struct NodeRegisterRequest {
pub node_id: NodeId,
pub listen_pg_addr: String,
pub listen_pg_port: u16,
pub listen_http_addr: String,
pub listen_http_port: u16,
pub availability_zone_id: AvailabilityZone,
}
#[derive(Serialize, Deserialize)]
pub struct NodeConfigureRequest {
pub node_id: NodeId,
pub availability: Option<NodeAvailabilityWrapper>,
pub scheduling: Option<NodeSchedulingPolicy>,
}
#[derive(Serialize, Deserialize)]
pub struct TenantPolicyRequest {
pub placement: Option<PlacementPolicy>,
pub scheduling: Option<ShardSchedulingPolicy>,
}
#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Hash)]
pub struct AvailabilityZone(pub String);
impl Display for AvailabilityZone {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
write!(f, "{}", self.0)
}
}
#[derive(Serialize, Deserialize)]
pub struct ShardsPreferredAzsRequest {
#[serde(flatten)]
pub preferred_az_ids: HashMap<TenantShardId, AvailabilityZone>,
}
#[derive(Serialize, Deserialize)]
pub struct ShardsPreferredAzsResponse {
pub updated: Vec<TenantShardId>,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct TenantLocateResponseShard {
pub shard_id: TenantShardId,
pub node_id: NodeId,
pub listen_pg_addr: String,
pub listen_pg_port: u16,
pub listen_http_addr: String,
pub listen_http_port: u16,
}
#[derive(Serialize, Deserialize)]
pub struct TenantLocateResponse {
pub shards: Vec<TenantLocateResponseShard>,
pub shard_params: ShardParameters,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct TenantDescribeResponse {
pub tenant_id: TenantId,
pub shards: Vec<TenantDescribeResponseShard>,
pub stripe_size: ShardStripeSize,
pub policy: PlacementPolicy,
pub config: TenantConfig,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct NodeShardResponse {
pub node_id: NodeId,
pub shards: Vec<NodeShard>,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct NodeShard {
pub tenant_shard_id: TenantShardId,
/// Whether the shard is observed secondary on a specific node. True = yes, False = no, None = not on this node.
pub is_observed_secondary: Option<bool>,
/// Whether the shard is intended to be a secondary on a specific node. True = yes, False = no, None = not on this node.
pub is_intended_secondary: Option<bool>,
}
#[derive(Serialize, Deserialize)]
pub struct NodeDescribeResponse {
pub id: NodeId,
pub availability: NodeAvailabilityWrapper,
pub scheduling: NodeSchedulingPolicy,
pub listen_http_addr: String,
pub listen_http_port: u16,
pub listen_pg_addr: String,
pub listen_pg_port: u16,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct TenantDescribeResponseShard {
pub tenant_shard_id: TenantShardId,
pub node_attached: Option<NodeId>,
pub node_secondary: Vec<NodeId>,
pub last_error: String,
/// A task is currently running to reconcile this tenant's intent state with the state on pageservers
pub is_reconciling: bool,
/// This shard failed in sending a compute notification to the cloud control plane, and a retry is pending.
pub is_pending_compute_notification: bool,
/// A shard split is currently underway
pub is_splitting: bool,
pub scheduling_policy: ShardSchedulingPolicy,
pub preferred_az_id: Option<String>,
}
/// Migration request for a given tenant shard to a given node.
///
/// Explicitly migrating a particular shard is a low level operation
/// TODO: higher level "Reschedule tenant" operation where the request
/// specifies some constraints, e.g. asking it to get off particular node(s)
#[derive(Serialize, Deserialize, Debug)]
pub struct TenantShardMigrateRequest {
pub tenant_shard_id: TenantShardId,
pub node_id: NodeId,
}
#[derive(Serialize, Clone, Debug)]
#[serde(into = "NodeAvailabilityWrapper")]
pub enum NodeAvailability {
// Normal, happy state
Active(PageserverUtilization),
// Node is warming up, but we expect it to become available soon. Covers
// the time span between the re-attach response being composed on the storage controller
// and the first successful heartbeat after the processing of the re-attach response
// finishes on the pageserver.
WarmingUp(Instant),
// Offline: Tenants shouldn't try to attach here, but they may assume that their
// secondary locations on this node still exist. Newly added nodes are in this
// state until we successfully contact them.
Offline,
}
impl PartialEq for NodeAvailability {
fn eq(&self, other: &Self) -> bool {
use NodeAvailability::*;
matches!(
(self, other),
(Active(_), Active(_)) | (Offline, Offline) | (WarmingUp(_), WarmingUp(_))
)
}
}
impl Eq for NodeAvailability {}
// This wrapper provides serde functionality and it should only be used to
// communicate with external callers which don't know or care about the
// utilisation score of the pageserver it is targeting.
#[derive(Serialize, Deserialize, Clone, Copy, Debug)]
pub enum NodeAvailabilityWrapper {
Active,
WarmingUp,
Offline,
}
impl From<NodeAvailabilityWrapper> for NodeAvailability {
fn from(val: NodeAvailabilityWrapper) -> Self {
match val {
// Assume the worst utilisation score to begin with. It will later be updated by
// the heartbeats.
NodeAvailabilityWrapper::Active => {
NodeAvailability::Active(PageserverUtilization::full())
}
NodeAvailabilityWrapper::WarmingUp => NodeAvailability::WarmingUp(Instant::now()),
NodeAvailabilityWrapper::Offline => NodeAvailability::Offline,
}
}
}
impl From<NodeAvailability> for NodeAvailabilityWrapper {
fn from(val: NodeAvailability) -> Self {
match val {
NodeAvailability::Active(_) => NodeAvailabilityWrapper::Active,
NodeAvailability::WarmingUp(_) => NodeAvailabilityWrapper::WarmingUp,
NodeAvailability::Offline => NodeAvailabilityWrapper::Offline,
}
}
}
#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
pub enum ShardSchedulingPolicy {
// Normal mode: the tenant's scheduled locations may be updated at will, including
// for non-essential optimization.
Active,
// Disable optimizations, but permit scheduling when necessary to fulfil the PlacementPolicy.
// For example, this still permits a node's attachment location to change to a secondary in
// response to a node failure, or to assign a new secondary if a node was removed.
Essential,
// No scheduling: leave the shard running wherever it currently is. Even if the shard is
// unavailable, it will not be rescheduled to another node.
Pause,
// No reconciling: we will make no location_conf API calls to pageservers at all. If the
// shard is unavailable, it stays that way. If a node fails, this shard doesn't get failed over.
Stop,
}
impl Default for ShardSchedulingPolicy {
fn default() -> Self {
Self::Active
}
}
#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
pub enum NodeSchedulingPolicy {
Active,
Filling,
Pause,
PauseForRestart,
Draining,
}
impl FromStr for NodeSchedulingPolicy {
type Err = anyhow::Error;
fn from_str(s: &str) -> Result<Self, Self::Err> {
match s {
"active" => Ok(Self::Active),
"filling" => Ok(Self::Filling),
"pause" => Ok(Self::Pause),
"pause_for_restart" => Ok(Self::PauseForRestart),
"draining" => Ok(Self::Draining),
_ => Err(anyhow::anyhow!("Unknown scheduling state '{s}'")),
}
}
}
impl From<NodeSchedulingPolicy> for String {
fn from(value: NodeSchedulingPolicy) -> String {
use NodeSchedulingPolicy::*;
match value {
Active => "active",
Filling => "filling",
Pause => "pause",
PauseForRestart => "pause_for_restart",
Draining => "draining",
}
.to_string()
}
}
/// Controls how tenant shards are mapped to locations on pageservers, e.g. whether
/// to create secondary locations.
#[derive(Clone, Serialize, Deserialize, Debug, PartialEq, Eq)]
pub enum PlacementPolicy {
/// Normal live state: one attached pageserver and zero or more secondaries.
Attached(usize),
/// Create one secondary mode locations. This is useful when onboarding
/// a tenant, or for an idle tenant that we might want to bring online quickly.
Secondary,
/// Do not attach to any pageservers. This is appropriate for tenants that
/// have been idle for a long time, where we do not mind some delay in making
/// them available in future.
Detached,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct TenantShardMigrateResponse {}
/// Metadata health record posted from scrubber.
#[derive(Serialize, Deserialize, Debug)]
pub struct MetadataHealthRecord {
pub tenant_shard_id: TenantShardId,
pub healthy: bool,
pub last_scrubbed_at: chrono::DateTime<chrono::Utc>,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct MetadataHealthUpdateRequest {
pub healthy_tenant_shards: HashSet<TenantShardId>,
pub unhealthy_tenant_shards: HashSet<TenantShardId>,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct MetadataHealthUpdateResponse {}
#[derive(Serialize, Deserialize, Debug)]
pub struct MetadataHealthListUnhealthyResponse {
pub unhealthy_tenant_shards: Vec<TenantShardId>,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct MetadataHealthListOutdatedRequest {
#[serde(with = "humantime_serde")]
pub not_scrubbed_for: Duration,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct MetadataHealthListOutdatedResponse {
pub health_records: Vec<MetadataHealthRecord>,
}
#[cfg(test)]
mod test {
use super::*;
use serde_json;
/// Check stability of PlacementPolicy's serialization
#[test]
fn placement_policy_encoding() -> anyhow::Result<()> {
let v = PlacementPolicy::Attached(1);
let encoded = serde_json::to_string(&v)?;
assert_eq!(encoded, "{\"Attached\":1}");
assert_eq!(serde_json::from_str::<PlacementPolicy>(&encoded)?, v);
let v = PlacementPolicy::Detached;
let encoded = serde_json::to_string(&v)?;
assert_eq!(encoded, "\"Detached\"");
assert_eq!(serde_json::from_str::<PlacementPolicy>(&encoded)?, v);
Ok(())
}
#[test]
fn test_reject_unknown_field() {
let id = TenantId::generate();
let create_request = serde_json::json!({
"new_tenant_id": id.to_string(),
"unknown_field": "unknown_value".to_string(),
});
let err = serde_json::from_value::<TenantCreateRequest>(create_request).unwrap_err();
assert!(
err.to_string().contains("unknown field `unknown_field`"),
"expect unknown field `unknown_field` error, got: {}",
err
);
}
}