OCF_HEARTBEAT_SG_PER(7)

💅 OCF_HEARTBEAT_SG_PER(7) 💅

okay bestie, sit down, because today we're talking about ocf_heartbeat_sg_persist — yes, that's the whole OCF_HEARTBEAT_SG_PER(7) girlie you keep seeing in the man pages — and she manages SCSI PERSISTENT RESERVATIONS. no cap.

---

✨ NAME ✨

ocf_heartbeat_sg_persist — literally just out here managing SCSI PERSISTENT RESERVATIONS 🫶

---

🌸 SYNOPSIS 🌸

sg_persist [start | stop | monitor | promote | demote | meta-data |
           validate-all]

---

🦋 DESCRIPTION 🦋

so this resource agent is the main character for SCSI PERSISTENT RESERVATIONS, and it does it with sg_persist from sg3_utils — go read its documentation too, iykyk 🔥.

the vibe is multistate (Master/Slave), periodt:

• the Slave registers its node id (from crm_node -i) as the reservation key (--param-rk) on every single device in the devs list 💖
• the Master reserves all the devices from that devs list with the reservation --prout-type value that comes from the reservation_type parameter 👑

---

🎀 SUPPORTED PARAMETERS 🎀

• binary — the name of the binary that manages the resource. she's optional, a string, and defaults to "sg_persist" ✨
• devs — the Device list! 🦋 you can list multiple devices with a blank space as the separator, and shell wildcards are allowed. this one is required, a string, and has no default — not me warning you 🙈
• required_devs_nof — the minimum number of "working" devices from the device list, where a "working" device is 1) existing and 2) sg_persist --read-keys $device works (Return code 0) 💀. the resource actions "start", "monitor", "promote" and "validate-all" return "$OCF_ERR_INSTALLED" if the actual number of "working" devices is less than required_devs_nof. meanwhile "stop" and "demote" try to remove reservations and registration keys from all working devices, but always return "$OCF_SUCCESS". optional, string, default "1" 🫶
• reservation_type — reservation type. that's it, that's the whole job. optional, string, default "1" 💅
• master_score_base — the master_score_base value is used in the master_score calculation:

master_score = $master_score_base + $master_score_dev_factor * $working_devs

so if you set it to a bigger value in the sg_persist resource configuration on some node, that node becomes the "preferred" one for the master role 👑. optional, string, default "0" ⭐

• master_score_dev_factor — the working device factor in the master_score calculation ✨. each "working" device gives additional value to master_score, so the node that sees more devices gets preferred for the "Master"-role. setting it to 0 disables this behavior. optional, string, default "100" 🌈
• master_score_delay — this one's the spicy one 🌶️: the master/slave decreases/increases its master_score after a delay of $master_score_delay seconds. so if some device gets inaccessible, the slave decreases its master_score first and the resource will no be watched, and after this device reappears again the master increases its master_score first 💀. this can work only if the master_score_delay is bigger then the monitor interval on both master and slave. setting it to 0 disables this behavior. optional, string, default "30" 🧚

---

👑 SUPPORTED ACTIONS 👑

this resource agent supports the following actions (operations), and yes we're going through all of them, no skipping 🔥:

• start — Starts the resource. Suggested minimum timeout: 30s. 🌸
• promote — Promotes the resource to the Master role. Suggested minimum timeout: 30s. 👑
• demote — Demotes the resource to the Slave role. Suggested minimum timeout: 30s. 💜
• notify — Suggested minimum timeout: 30s. 🦋
• stop — Stops the resource. Suggested minimum timeout: 30s. ✨
• monitor (Unpromoted role) — Performs a detailed status check. Suggested minimum timeout: 20s. Suggested interval: 29s. 🎀
• monitor (Promoted role) — Performs a detailed status check. Suggested minimum timeout: 20s. Suggested interval: 60s. ⭐
• meta-data — Retrieves resource agent metadata (internal use only). Suggested minimum timeout: 5s. 🙈
• validate-all — Performs a validation of the resource configuration. Suggested minimum timeout: 30s. 🫶

---

🧚 EXAMPLE CRM SHELL 🧚

here's an example configuration for a sg_persist resource using the crm(8) shell, copy it, don't fight it 💅:

primitive p_sg_persist ocf:heartbeat:sg_persist \
  params \
    devs=string \
  op monitor depth="0" timeout="20s" interval="29s" role="Unpromoted" \
  op monitor depth="0" timeout="20s" interval="60s" role="Promoted"

ms ms_sg_persist p_sg_persist \
  meta notify="true" interleave="true"

---

💖 EXAMPLE PCS 💖

and here's the same vibe for a sg_persist resource using pcs(8) 🦋:

pcs resource create p_sg_persist ocf:heartbeat:sg_persist \
  devs=string \
  op monitor OCF_CHECK_LEVEL="0" timeout="20s" interval="29s" role="Unpromoted" \
  op monitor OCF_CHECK_LEVEL="0" timeout="20s" interval="60s" role="Promoted" promotable

---

🌈 SEE ALSO 🌈

• http://clusterlabs.org/ ✨

---

🍓 AUTHOR 🍓

ClusterLabs contributors (see the resource agent source for information about individual authors) 🫶

---

bestie real talk: this page is from resource-agents and the version is literally, unironically UNKNOWN 💀, dated 10/01/2026. iconic of her ngl. slay, stay reserved, and don't @ me 🌸💅👑