use serde::{Deserialize, Serialize}; use proxmox::const_regex; use proxmox::api::{api, schema::*}; use crate::{ Userid, Authid, REMOTE_ID_SCHEMA, DRIVE_NAME_SCHEMA, MEDIA_POOL_NAME_SCHEMA, SINGLE_LINE_COMMENT_SCHEMA, PROXMOX_SAFE_ID_FORMAT, DATASTORE_SCHEMA, }; const_regex!{ /// Regex for verification jobs 'DATASTORE:ACTUAL_JOB_ID' pub VERIFICATION_JOB_WORKER_ID_REGEX = concat!(r"^(", PROXMOX_SAFE_ID_REGEX_STR!(), r"):"); /// Regex for sync jobs 'REMOTE:REMOTE_DATASTORE:LOCAL_DATASTORE:ACTUAL_JOB_ID' pub SYNC_JOB_WORKER_ID_REGEX = concat!(r"^(", PROXMOX_SAFE_ID_REGEX_STR!(), r"):(", PROXMOX_SAFE_ID_REGEX_STR!(), r"):(", PROXMOX_SAFE_ID_REGEX_STR!(), r"):"); } pub const UPID_SCHEMA: Schema = StringSchema::new("Unique Process/Task ID.") .max_length(256) .schema(); pub const JOB_ID_SCHEMA: Schema = StringSchema::new("Job ID.") .format(&PROXMOX_SAFE_ID_FORMAT) .min_length(3) .max_length(32) .schema(); pub const SYNC_SCHEDULE_SCHEMA: Schema = StringSchema::new( "Run sync job at specified schedule.") .format(&ApiStringFormat::VerifyFn(pbs_systemd::time::verify_calendar_event)) .type_text("") .schema(); pub const GC_SCHEDULE_SCHEMA: Schema = StringSchema::new( "Run garbage collection job at specified schedule.") .format(&ApiStringFormat::VerifyFn(pbs_systemd::time::verify_calendar_event)) .type_text("") .schema(); pub const PRUNE_SCHEDULE_SCHEMA: Schema = StringSchema::new( "Run prune job at specified schedule.") .format(&ApiStringFormat::VerifyFn(pbs_systemd::time::verify_calendar_event)) .type_text("") .schema(); pub const VERIFICATION_SCHEDULE_SCHEMA: Schema = StringSchema::new( "Run verify job at specified schedule.") .format(&ApiStringFormat::VerifyFn(pbs_systemd::time::verify_calendar_event)) .type_text("") .schema(); pub const REMOVE_VANISHED_BACKUPS_SCHEMA: Schema = BooleanSchema::new( "Delete vanished backups. This remove the local copy if the remote backup was deleted.") .default(true) .schema(); #[api( properties: { "next-run": { description: "Estimated time of the next run (UNIX epoch).", optional: true, type: Integer, }, "last-run-state": { description: "Result of the last run.", optional: true, type: String, }, "last-run-upid": { description: "Task UPID of the last run.", optional: true, type: String, }, "last-run-endtime": { description: "Endtime of the last run.", optional: true, type: Integer, }, } )] #[derive(Serialize,Deserialize,Default)] #[serde(rename_all="kebab-case")] /// Job Scheduling Status pub struct JobScheduleStatus { #[serde(skip_serializing_if="Option::is_none")] pub next_run: Option, #[serde(skip_serializing_if="Option::is_none")] pub last_run_state: Option, #[serde(skip_serializing_if="Option::is_none")] pub last_run_upid: Option, #[serde(skip_serializing_if="Option::is_none")] pub last_run_endtime: Option, } #[api()] #[derive(Debug, Copy, Clone, PartialEq, Serialize, Deserialize)] #[serde(rename_all = "lowercase")] /// When do we send notifications pub enum Notify { /// Never send notification Never, /// Send notifications for failed and successful jobs Always, /// Send notifications for failed jobs only Error, } #[api( properties: { gc: { type: Notify, optional: true, }, verify: { type: Notify, optional: true, }, sync: { type: Notify, optional: true, }, }, )] #[derive(Debug, Serialize, Deserialize)] /// Datastore notify settings pub struct DatastoreNotify { /// Garbage collection settings pub gc: Option, /// Verify job setting pub verify: Option, /// Sync job setting pub sync: Option, } pub const DATASTORE_NOTIFY_STRING_SCHEMA: Schema = StringSchema::new( "Datastore notification setting") .format(&ApiStringFormat::PropertyString(&DatastoreNotify::API_SCHEMA)) .schema(); pub const IGNORE_VERIFIED_BACKUPS_SCHEMA: Schema = BooleanSchema::new( "Do not verify backups that are already verified if their verification is not outdated.") .default(true) .schema(); pub const VERIFICATION_OUTDATED_AFTER_SCHEMA: Schema = IntegerSchema::new( "Days after that a verification becomes outdated") .minimum(1) .schema(); #[api( properties: { id: { schema: JOB_ID_SCHEMA, }, store: { schema: DATASTORE_SCHEMA, }, "ignore-verified": { optional: true, schema: IGNORE_VERIFIED_BACKUPS_SCHEMA, }, "outdated-after": { optional: true, schema: VERIFICATION_OUTDATED_AFTER_SCHEMA, }, comment: { optional: true, schema: SINGLE_LINE_COMMENT_SCHEMA, }, schedule: { optional: true, schema: VERIFICATION_SCHEDULE_SCHEMA, }, } )] #[derive(Serialize,Deserialize)] #[serde(rename_all="kebab-case")] /// Verification Job pub struct VerificationJobConfig { /// unique ID to address this job pub id: String, /// the datastore ID this verificaiton job affects pub store: String, #[serde(skip_serializing_if="Option::is_none")] /// if not set to false, check the age of the last snapshot verification to filter /// out recent ones, depending on 'outdated_after' configuration. pub ignore_verified: Option, #[serde(skip_serializing_if="Option::is_none")] /// Reverify snapshots after X days, never if 0. Ignored if 'ignore_verified' is false. pub outdated_after: Option, #[serde(skip_serializing_if="Option::is_none")] pub comment: Option, #[serde(skip_serializing_if="Option::is_none")] /// when to schedule this job in calendar event notation pub schedule: Option, } #[api( properties: { config: { type: VerificationJobConfig, }, status: { type: JobScheduleStatus, }, }, )] #[derive(Serialize,Deserialize)] #[serde(rename_all="kebab-case")] /// Status of Verification Job pub struct VerificationJobStatus { #[serde(flatten)] pub config: VerificationJobConfig, #[serde(flatten)] pub status: JobScheduleStatus, } #[api( properties: { store: { schema: DATASTORE_SCHEMA, }, pool: { schema: MEDIA_POOL_NAME_SCHEMA, }, drive: { schema: DRIVE_NAME_SCHEMA, }, "eject-media": { description: "Eject media upon job completion.", type: bool, optional: true, }, "export-media-set": { description: "Export media set upon job completion.", type: bool, optional: true, }, "latest-only": { description: "Backup latest snapshots only.", type: bool, optional: true, }, "notify-user": { optional: true, type: Userid, }, } )] #[derive(Serialize,Deserialize,Clone)] #[serde(rename_all="kebab-case")] /// Tape Backup Job Setup pub struct TapeBackupJobSetup { pub store: String, pub pool: String, pub drive: String, #[serde(skip_serializing_if="Option::is_none")] pub eject_media: Option, #[serde(skip_serializing_if="Option::is_none")] pub export_media_set: Option, #[serde(skip_serializing_if="Option::is_none")] pub latest_only: Option, /// Send job email notification to this user #[serde(skip_serializing_if="Option::is_none")] pub notify_user: Option, } #[api( properties: { id: { schema: JOB_ID_SCHEMA, }, setup: { type: TapeBackupJobSetup, }, comment: { optional: true, schema: SINGLE_LINE_COMMENT_SCHEMA, }, schedule: { optional: true, schema: SYNC_SCHEDULE_SCHEMA, }, } )] #[derive(Serialize,Deserialize,Clone)] #[serde(rename_all="kebab-case")] /// Tape Backup Job pub struct TapeBackupJobConfig { pub id: String, #[serde(flatten)] pub setup: TapeBackupJobSetup, #[serde(skip_serializing_if="Option::is_none")] pub comment: Option, #[serde(skip_serializing_if="Option::is_none")] pub schedule: Option, } #[api( properties: { config: { type: TapeBackupJobConfig, }, status: { type: JobScheduleStatus, }, }, )] #[derive(Serialize,Deserialize)] #[serde(rename_all="kebab-case")] /// Status of Tape Backup Job pub struct TapeBackupJobStatus { #[serde(flatten)] pub config: TapeBackupJobConfig, #[serde(flatten)] pub status: JobScheduleStatus, /// Next tape used (best guess) #[serde(skip_serializing_if="Option::is_none")] pub next_media_label: Option, } #[api( properties: { id: { schema: JOB_ID_SCHEMA, }, store: { schema: DATASTORE_SCHEMA, }, "owner": { type: Authid, optional: true, }, remote: { schema: REMOTE_ID_SCHEMA, }, "remote-store": { schema: DATASTORE_SCHEMA, }, "remove-vanished": { schema: REMOVE_VANISHED_BACKUPS_SCHEMA, optional: true, }, comment: { optional: true, schema: SINGLE_LINE_COMMENT_SCHEMA, }, schedule: { optional: true, schema: SYNC_SCHEDULE_SCHEMA, }, } )] #[derive(Serialize,Deserialize,Clone,Updater)] #[serde(rename_all="kebab-case")] /// Sync Job pub struct SyncJobConfig { #[updater(skip)] pub id: String, pub store: String, #[serde(skip_serializing_if="Option::is_none")] pub owner: Option, pub remote: String, pub remote_store: String, #[serde(skip_serializing_if="Option::is_none")] pub remove_vanished: Option, #[serde(skip_serializing_if="Option::is_none")] pub comment: Option, #[serde(skip_serializing_if="Option::is_none")] pub schedule: Option, } #[api( properties: { config: { type: SyncJobConfig, }, status: { type: JobScheduleStatus, }, }, )] #[derive(Serialize,Deserialize)] #[serde(rename_all="kebab-case")] /// Status of Sync Job pub struct SyncJobStatus { #[serde(flatten)] pub config: SyncJobConfig, #[serde(flatten)] pub status: JobScheduleStatus, }