WARNING

このページはまだ翻訳されていません。英語で表示しています。

Sites

A site is one Arkvory installation: the main one, a mirror in another location, an installation that imports versions from another. On サイト, one installation, the controller, shows every site it manages: whether it answers, its version, its HA copies and read gateways, its mirrors and how far behind they are, and its download queue. Only global administrators see the page.

State of the cluster

The page starts with クラスターの状態: what needs attention across the sites, and which site feeds which. The controller builds it from the last reads of the sites, at most once every 5 seconds, and lists itself as "this site".

Alerts

Each alert names the site and the problem. The most severe come first: 重大, then 警告, then 情報. When nothing needs attention, the list says so.

Alert	Severity	When	What to do
Unreachable	重大	No answer for 2 minutes. One failed read is no alert	Check the site and the network
Certificate, revoked, refused, invalid	重大	At once, with the state of the same name	Pair again, or update both installations
Stale	警告	The site answered, but was not read for three read periods (90 seconds) plus the time one round of reads takes: the background reads of the controller are stuck	Check the API process of the controller
Not writable	重大	A writer refuses writes and is not draining	Check its database and its disk
Copies	重大	An HA site lacks copies of its data; writes pause until they are back	Bring the missing node back, see HA cluster
Single copy	警告	Copies are lacking, but an operator accepted one copy until a set time	Bring the missing node back before that time
Mirror failing	警告	The last synchronization of a mirror failed. Reading a restored source again does not count	Read the error code of the mirror
Mirror behind	警告	A mirror is ARKVORY_FLEET_LAG_ALERT changes (1000 by default) or more behind its source	Check the source, the network and the mirror
Updates failed	警告	The updater of the site reports a failure	Open 更新 in the settings of the site
Version skew	情報	The site runs another version than the controller	Update the sites to the same release

Alerts use only fresh reads: a site that does not answer raises its own alert, not alerts about what it last reported. One round of reads takes up to 15 seconds for every eight sites, because the controller reads at most eight at a time and waits up to 15 seconds for each; the stale alert adds that time, so many unreachable sites do not make healthy ones look stuck. Prometheus alerts at the same ARKVORY_FLEET_LAG_ALERT, see Monitoring.

Links

リンク shows which site feeds which. Every mirror and import is an arrow from the source site to the site that holds the mirror.

• A source at whose address no managed site answers is drawn by its address, with a dashed border. If that source is the controller itself, set ARKVORY_SITE_URL, see Settings.
• A failing link is dashed red.
• The number on a link is how many changes the mirror is behind its source.
• While a site does not answer, its links and their lag stay as it last reported them, marked as last known: dashed on the map, and with the hint 最後に確認した状態: サイトが応答しません in the table.

Below the map, a table lists the same facts: コピー元, サイト, リポジトリ, コピー元のリポジトリ and 遅れ.

Global administrators of the controller get the same alerts and links from GET /api/v1/fleet/overview, or from the SDK with client.fleet.overview(). For Prometheus, the controller exports the state of its sites and the number of alerts: see Monitoring.

How it works

• The controller keeps a list of sites. It reads every site every 30 seconds over HTTPS and keeps the last answer, so a site that stops answering still shows its last known state.
• When a site is paired, it gives the controller a key of its own, a fleet key. The fleet key reaches only what a controller needs: the state of the site, the descriptions in its catalog, and the listed changes: the download queue, update requests, and mirror and HA changes for the site's agent. It cannot download file bytes, write to repositories, or read or change accounts and keys.
• The controller keeps the fleet keys sealed with a key in its data directory (fleet.key). The database and its backups alone do not reveal them. After a restore, pair the sites again.
• Every site shows which controllers manage it and can revoke them at any time.

Pair a site

1. On the site: open サイト and, under このサイト, select ペアリングコードを作成. The site shows a code that is valid for ten minutes and, with built-in HTTPS, the fingerprint of its certificate.
2. On the controller: open サイト, enter the address of the site under サイトをペアリング and select 確認. The controller connects and shows the fingerprint of the certificate it received.
3. Compare the two fingerprints. They must be the same. If they differ, the address is wrong or something stands between the two installations: do not pair.
4. Enter the code and select ペアリング. The site appears in the list with its state.

A code works once. A used or expired code is refused: make a new one.

The address is the HTTPS address of the site's API, for example https://branch.example:8443. Plain HTTP is accepted only on the same computer (127.0.0.1). When TLS ends at a proxy in front of the site, the site cannot show the proxy's certificate: compare with the SHA-256 fingerprint your browser shows for the site's address.

From then on, the controller trusts exactly that certificate. When the site gets a new certificate, it shows 証明書が変更: remove the site and pair it again.

States

State	Meaning	What to do
稼働中	The site answered the last read	Nothing
未読み取り	The site was not read yet	Wait for the next read or select 今すぐ読む
到達不能	No answer: the site is stopped, or the network is down	Check the site and the network. The last known state stays visible
証明書が変更	The site presents another certificate than the one confirmed	If the certificate was renewed, remove the site and pair it again
アクセス取り消し	The administrator of the site revoked this controller	Pair again with a new code
想定外の応答	The site answered in a way this release does not understand	Update both installations to the same release
拒否	The site refused the request	Pair again with a new code

What the list shows

• バージョン: the release of the site and its role, writer or read gateway.
• HA とゲートウェイ: on an HA cluster, the complete copies and the copies a write needs; the read gateways that hold their slot.
• ミラー: the mirrors and imports of the site; how many changes the furthest one is behind its source, or how many are stopped.
• ダウンロードキュー: downloads that run and downloads that wait in the download queue.
• 最後の応答: when the site last answered.

すべて今すぐ読む reads every site now instead of waiting for the next round. 削除 forgets a site here; revoke the controller on the site as well, because until then its key stays valid there.

Content of a site

コンテンツ in the row of a site opens its repositories, artifacts and files below the list. Choose a repository; 検索 searches artifacts by name or label and lists files under a path that starts with what you typed. さらに表示 reads the next page.

The controller reads these on the site with its fleet key and shows the answer as the site gave it. Only descriptions travel: names, sizes, labels, stages, paths and revisions. To download a file, use the site itself.

Settings of a site

設定 in the row of a site opens its settings below the list; このサイトの設定 opens the same card for this site. The card has three parts:

• ダウンロードキュー: how many downloads run at once, places kept for people, the strategy, class weights and the other queue values of Traffic control. An empty field keeps the value of the site's environment, shown in grey. キューの設定を保存 applies the values at once, without a restart: new decisions follow them, active downloads keep their places. 既定値に戻す returns to the environment.
• 更新: the installed and the latest version, automatic installation and the maintenance hour, 今すぐ確認 and 更新をインストール. The site's updater takes a request within a minute; the card shows it as accepted until then.
• ミラーとクラスター: the site's mirrors and imports with キーを置き換える and 切り離す, ミラーまたはインポートを接続 with the source key, and on an HA cluster 切り替える to a node with a complete copy (the node name typed twice) and the single-copy decision. The site's agent applies each request within a minute with the checks and rollback of arkvory configure and cluster-*; エージェントへのリクエスト shows how each ended. The source key goes to the agent once, in a file of its own, and is never shown again. The agent is the updater of the site: connect it with updates-connect.

Each change carries the revision it was made from. If someone changed the settings meanwhile, the site refuses the change and the card shows the current values: check them and save again. The site records who changed its queue: the controller's key and the name of its administrator.

From the controller a mirror or an import is attached only as a new repository, and the authority certificate of a source is set on the site itself (arkvory configure --mirror-ca-file): what a site trusts never changes remotely. The request list shows how each change ended as a code; the agent's own words stay in its log on the site. If a change was interrupted half done, the agent takes no other until an administrator of the site checked it and ran arkvory changes-reset --root <root> there.

Who manages this site

Under このサイト, every site lists the controllers that manage it: their name, who paired them, when, and their last request. 取り消す refuses the controller's key at once.

Settings

Variable	Default	Meaning
ARKVORY_SITE_NAME	the host name	How this installation is named on other sites
ARKVORY_FLEET_KEY_FILE	fleet.key in the data directory	The key that seals the keys of managed sites; made on the first start of the writer
ARKVORY_SITE_URL	not set	The https://host[:port] address other sites use to reach this installation. With it, mirrors whose source is this installation link to it on リンク. An invalid value stops the start
ARKVORY_FLEET_LAG_ALERT	1000	Changes behind its source at which a mirror of a managed site alerts; a whole number, 1 or more

Limits

• Fencing tests and adding or removing cluster nodes stay local commands on the nodes. Read gateways keep the download queue of their environment.
• A controller manages up to 256 sites and reads at most eight at a time.
• The controller holds a key to every site it manages: protect its global administrators and its data directory as carefully as the sites themselves.